Databases - JB on programming
Theoretical Introduction
Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Terminology and overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
General-purpose and special-purpose DBMSs . . . . . . . . . . . . . . . . . . . . . . . .
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Design and modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Performance, security, and availability . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1.10 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1.11 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1.12 Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.1.13 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Schema migration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Risks and Benefits
Schema migration in agile software development
. . . . . . . . . . . . . . . . . . . . . .
Available Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Star schema . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Not Only SQL
CAP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Science and medicine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Organisations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Companies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Projects, programs, policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Military . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Certifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Other . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Eventual consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Conflict resolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Strong eventual consistency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Object-relational impedance mismatch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mismatches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Solving impedance mismatch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Contention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Philosophical differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Object database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Timeline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Adoption of object databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Technical features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Standards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Comparison with RDBMSs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.10 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
NoSQL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Types and examples of NoSQL databases . . . . . . . . . . . . . . . . . . . . . . . . . .
Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Handling relational data
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ACID and JOIN Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Key-value database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Types and notable examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Document-oriented database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Comparison with relational databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Implementations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
NewSQL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ACID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Consistency (database systems) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
As an ACID guarantee . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
As a CAP trade-off . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Durability (database systems) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Serializability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Database transaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Correctness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
View and conflict serializability
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Enforcing conflict serializability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Distributed serializability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Isolation (database systems) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Concurrency control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Isolation levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Default isolation level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Read phenomena . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Isolation Levels, Read Phenomena and Locks . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Database transaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Transactional databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Object databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Distributed transactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Transactional filesystems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Transaction processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ACID criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Implementations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Journaling file system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Rationale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Alternatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Atomicity (database systems) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Lock (database) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Mechanisms for locking
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Record locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Granularity of locks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Use of locks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Two-phase locking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Data-access locks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Two-phase locking and its special cases . . . . . . . . . . . . . . . . . . . . . . . . . . .
Deadlocks in 2PL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Multiversion concurrency control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Version control systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Snapshot isolation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Workarounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Two-phase commit protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Basic algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Disadvantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Implementing the two-phase commit protocol . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Three-phase commit protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Protocol Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Disadvantages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Horizontal and vertical scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Database scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Strong versus eventual consistency (storage) . . . . . . . . . . . . . . . . . . . . . . . . .
Performance tuning versus hardware scalability . . . . . . . . . . . . . . . . . . . . . . .
Weak versus strong scaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8.1.10 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Shard (database architecture) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Database architecture
Shards compared to horizontal partitioning
Support for shards
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Disadvantages of sharding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Etymology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Optimistic concurrency control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
OCC phases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Web usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Partition (database) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Benefits of multiple partitions
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Partitioning criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Partitioning methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Distributed transaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Redis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Supported languages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Data types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Persistence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Replication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
MongoDB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Main features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Criticisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Production deployments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.2.10 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PostgreSQL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Name . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Multiversion concurrency control (MVCC)
. . . . . . . . . . . . . . . . . . . . . . . . .
Storage and replication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Control and connectivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Security . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Upcoming features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Benchmarks and performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3.10 Platforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3.11 Database administration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3.12 Prominent users
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9.3.13 Proprietary derivatives and support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9.3.14 Release history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9.3.15 See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9.3.16 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9.3.17 Further reading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
9.3.18 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Apache Cassandra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Licensing and support
Main features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Data model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Prominent users
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
9.4.10 External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
Berkeley DB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Editions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Programs that use Berkeley DB
Licensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Memcached . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Software architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
Example code
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
BigTable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Other similar software . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
See also . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
External links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
10 Text and image sources, contributors, and licenses
10.1 Text . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
10.2 Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
10.3 Content license . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Chapter 1
1.1 Database
Because of the close relationship between them, the
term “database” is often used casually to refer to both a
“Database Software” redirects here. For the computer database and the DBMS used to manipulate it.
program, see Europress.
Outside the world of professional information technology, the term database is often used to refer to any collec[1]
A database is an organized collection of data. It is the tion of related data (such as a spreadsheet or a card index).
collection of schemas, tables, queries, reports, views and This article is concerned only with databases where the
necessitate use of a database
other objects. The data is typically organized to model as- size and usage requirements
pects of reality in a way that supports processes requiring
information, such as modelling the availability of rooms Existing DBMSs provide various functions that allow
in hotels in a way that supports finding a hotel with va- management of a database and its data which can be clascancies.
sified into four main functional groups:
A database management system (DBMS) is a
computer software application that interacts with the
user, other applications, and the database itself to capture and analyze data. A general-purpose DBMS is designed to allow the definition, creation, querying, update,
and administration of databases. Well-known DBMSs
include MySQL, PostgreSQL, Microsoft SQL Server,
Oracle, Sybase and IBM DB2. A database is not generally
portable across different DBMSs, but different DBMS
can interoperate by using standards such as SQL and
ODBC or JDBC to allow a single application to work with
more than one DBMS. Database management systems
are often classified according to the database model that
they support; the most popular database systems since
the 1980s have all supported the relational model as represented by the SQL language. Sometimes a DBMS is
loosely referred to as a 'database'.
• Data definition – Creation, modification and removal
of definitions that define the organization of the data.
• Update – Insertion, modification, and deletion of the
actual data.[3]
• Retrieval – Providing information in a form directly
usable or for further processing by other applications. The retrieved data may be made available in a
form basically the same as it is stored in the database
or in a new form obtained by altering or combining
existing data from the database.[4]
• Administration – Registering and monitoring users,
enforcing data security, monitoring performance,
maintaining data integrity, dealing with concurrency
control, and recovering information that has been
corrupted by some event such as an unexpected system failure.[5]
Terminology and overview
Both a database and its DBMS conform to the princi[6]
Formally, a “database” refers to a set of related data and ples of a particular database model. “Database system”
model, database manthe way it is organized. Access to this data is usually pro- refers collectively to the database
vided by a “database management system” (DBMS) consisting of an integrated set of computer software that al- Physically, database servers are dedicated computers
lows users to interact with one or more databases and pro- that hold the actual databases and run only the DBMS
vides access to all of the data contained in the database and related software. Database servers are usually
(although restrictions may exist that limit access to par- multiprocessor computers, with generous memory and
ticular data). The DBMS provides various functions that RAID disk arrays used for stable storage. RAID is used
allow entry, storage and retrieval of large quantities of for recovery of data if any of the disks fail. Hardware
information and provides ways to manage how that infor- database accelerators, connected to one or more servers
via a high-speed channel, are also used in large volume
mation is organized.
transaction processing environments. DBMSs are found
at the heart of most database applications. DBMSs may
be built around a custom multitasking kernel with builtin networking support, but modern DBMSs typically rely
on a standard operating system to provide these functions.
Since DBMSs comprise a significant economical market,
computer and storage vendors often take into account
DBMS requirements in their own development plans.
cesses the database on behalf of end-users, without exposing the DBMS interface directly. Application programmers may use a wire protocol directly, or more likely
through an application programming interface. Database
designers and database administrators interact with the
DBMS through dedicated interfaces to build and maintain the applications’ databases, and thus need some more
knowledge and understanding about how DBMSs operate
and the DBMSs’ external interfaces and tuning parameDatabases and DBMSs can be categorized according to
the database model(s) that they support (such as relational ters.
or XML), the type(s) of computer they run on (from a
server cluster to a mobile phone), the query language(s)
used to access the database (such as SQL or XQuery), and
their internal engineering, which affects performance,
1.1.4 History
scalability, resilience, and security.
Databases are used to support internal operations of organizations and to underpin online interactions with customers and suppliers (see Enterprise software).
Databases are used to hold administrative information
and more specialized data, such as engineering data or
economic models. Examples of database applications
include computerized library systems, flight reservation
systems, computerized parts inventory systems, and many
content management systems that store websites as collections of webpages in a database.
purpose DBMSs
A DBMS has evolved into a complex software system and
its development typically requires thousands of personyears of development effort.[8] Some general-purpose
DBMSs such as Adabas, Oracle and DB2 have been
undergoing upgrades since the 1970s. General-purpose
DBMSs aim to meet the needs of as many applications
as possible, which adds to the complexity. However, the
fact that their development cost can be spread over a large
number of users means that they are often the most costeffective approach. However, a general-purpose DBMS
is not always the optimal solution: in some cases a
general-purpose DBMS may introduce unnecessary overhead. Therefore, there are many examples of systems that
use special-purpose databases. A common example is an
email system that performs many of the functions of a
general-purpose DBMS such as the insertion and deletion of messages composed of various items of data or
associating messages with a particular email address; but
these functions are limited to what is required to handle
email and don't provide the user with the all of the functionality that would be available using a general-purpose
Following the technology progress in the areas of
processors, computer memory, computer storage and
computer networks, the sizes, capabilities, and performance of databases and their respective DBMSs
have grown in orders of magnitude. The development of database technology can be divided into three
eras based on data model or structure: navigational,[9]
SQL/relational, and post-relational.
The two main early navigational data models were the
hierarchical model, epitomized by IBM’s IMS system,
and the CODASYL model (network model), implemented in a number of products such as IDMS.
The relational model, first proposed in 1970 by Edgar F.
Codd, departed from this tradition by insisting that applications should search for data by content, rather than
by following links. The relational model employs sets
of ledger-style tables, each used for a different type of
entity. Only in the mid-1980s did computing hardware
become powerful enough to allow the wide deployment
of relational systems (DBMSs plus applications). By the
early 1990s, however, relational systems dominated in all
large-scale data processing applications, and as of 2015
they remain dominant : IBM DB2, Oracle, MySQL and
Microsoft SQL Server are the top DBMS.[10] The dominant database language, standardised SQL for the relational model, has influenced database languages for other
data models.
Object databases were developed in the 1980s to overcome the inconvenience of object-relational impedance
mismatch, which led to the coining of the term “postrelational” and also the development of hybrid objectrelational databases.
The next generation of post-relational databases in the
late 2000s became known as NoSQL databases, introducing fast key-value stores and document-oriented
databases. A competing “next generation” known as
NewSQL databases attempted new implementations that
retained the relational/SQL model while aiming to match
the high performance of NoSQL compared to commerMany other databases have application software that ac- cially available relational DBMSs.
1960s, navigational DBMS
Further information: Navigational database
The introduction of the term database coincided with
gation of a linked data set which was formed into a large
network. Applications could find records by one of three
1. Use of a primary key (known as a CALC key, typically implemented by hashing)
2. Navigating relationships (called sets) from one
record to another
3. Scanning all the records in a sequential order
Later systems added B-trees to provide alternate access
paths. Many CODASYL databases also added a very
straightforward query language. However, in the final
tally, CODASYL was very complex and required significant training and effort to produce useful applications.
IBM also had their own DBMS in 1968, known as
Information Management System (IMS). IMS was a development of software written for the Apollo program
on the System/360. IMS was generally similar in concept to CODASYL, but used a strict hierarchy for its
model of data navigation instead of CODASYL’s network model. Both concepts later became known as navigational databases due to the way data was accessed, and
Bachman’s 1973 Turing Award presentation was The Programmer as Navigator. IMS is classified as a hierarchical
database. IDMS and Cincom Systems' TOTAL database
are classified as network databases. IMS remains in use
as of 2014.[12]
1970s, relational DBMS
Basic structure of navigational CODASYL database model
Edgar Codd worked at IBM in San Jose, California, in
one of their offshoot offices that was primarily involved
in the development of hard disk systems. He was unhappy with the navigational model of the CODASYL approach, notably the lack of a “search” facility. In 1970, he
wrote a number of papers that outlined a new approach
to database construction that eventually culminated in the
groundbreaking A Relational Model of Data for Large
Shared Data Banks.[13]
the availability of direct-access storage (disks and drums)
from the mid-1960s onwards. The term represented a
contrast with the tape-based systems of the past, allowing
shared interactive use rather than daily batch processing.
The Oxford English Dictionary cites[11] a 1962 report by
the System Development Corporation of California as the
first to use the term “data-base” in a specific technical
In this paper, he described a new system for storing and
working with large databases. Instead of records being
As computers grew in speed and capability, a number of stored in some sort of linked list of free-form records as
general-purpose database systems emerged; by the mid- in CODASYL, Codd’s idea was to use a "table" of fixed1960s a number of such systems had come into commer- length records, with each table used for a different type
cial use. Interest in a standard began to grow, and Charles of entity. A linked-list system would be very inefficient
Bachman, author of one such product, the Integrated Data when storing “sparse” databases where some of the data
Store (IDS), founded the “Database Task Group” within for any one record could be left empty. The relational
CODASYL, the group responsible for the creation and model solved this by splitting the data into a series of
standardization of COBOL. In 1971 the Database Task normalized tables (or relations), with optional elements
Group delivered their standard, which generally became being moved out of the main table to where they would
known as the “CODASYL approach”, and soon a number take up room only if needed. Data may be freely inserted,
of commercial products based on this approach entered deleted and edited in these tables, with the DBMS doing
the market.
whatever maintenance needed to present a table view to
The CODASYL approach relied on the “manual” navi- the application/user.
login first
mark Samuel Clemens
Amber Straub
login phone
mark 555.555.5555
"related table"
In the relational model, records are “linked” using virtual keys
not stored in the database but defined as needed between the data
contained in the records.
later spawn the ubiquitous SQL. Using a branch of mathematics known as tuple calculus, he demonstrated that
such a system could support all the operations of normal
databases (inserting, updating etc.) as well as providing
a simple system for finding and returning sets of data in a
single operation.
Codd’s paper was picked up by two people at Berkeley,
Eugene Wong and Michael Stonebraker. They started
a project known as INGRES using funding that had already been allocated for a geographical database project
and student programmers to produce code. Beginning
in 1973, INGRES delivered its first test products which
were generally ready for widespread use in 1979. INGRES was similar to System R in a number of ways, including the use of a “language” for data access, known as
QUEL. Over time, INGRES moved to the emerging SQL
IBM itself did one test implementation of the relational
model, PRTV, and a production one, Business System
12, both now discontinued. Honeywell wrote MRDS
for Multics, and now there are two new implementaThe relational model also allowed the content of the
tions: Alphora Dataphor and Rel. Most other DBMS imdatabase to evolve without constant rewriting of links and
plementations usually called relational are actually SQL
pointers. The relational part comes from entities referDBMSs.
encing other entities in what is known as one-to-many
relationship, like a traditional hierarchical model, and In 1970, the University of Michigan began developmany-to-many relationship, like a navigational (network) ment of the MICRO Information Management Sysmodel. Thus, a relational model can express both hierar- tem[14] based on D.L. Childs’ Set-Theoretic Data
chical and navigational models, as well as its native tab- model.[15][16][17] Micro was used to manage very large
ular model, allowing for pure or combined modeling in data sets by the US Department of Labor, the U.S. Enviterms of these three models, as the application requires. ronmental Protection Agency, and researchers from the
University of Alberta, the University of Michigan, and
For instance, a common use of a database system is to
Wayne State University. It ran on IBM mainframe comtrack information about users, their name, login informaputers using the Michigan Terminal System.[18] The systion, various addresses and phone numbers. In the navtem remained in production until 1998.
igational approach all of these data would be placed in
a single record, and unused items would simply not be
placed in the database. In the relational approach, the
data would be normalized into a user table, an address Integrated approach
table and a phone number table (for instance). Records
would be created in these optional tables only if the ad- Main article: Database machine
dress or phone numbers were actually provided.
Linking the information back together is the key to this
system. In the relational model, some bit of information
was used as a "key", uniquely defining a particular record.
When information was being collected about a user, information stored in the optional tables would be found by
searching for this key. For instance, if the login name of a
user is unique, addresses and phone numbers for that user
would be recorded with the login name as its key. This
simple “re-linking” of related data back into a single collection is something that traditional computer languages
are not designed for.
Just as the navigational approach would require programs
to loop in order to collect records, the relational approach would require loops to collect information about
any one record. Codd’s solution to the necessary looping was a set-oriented language, a suggestion that would
In the 1970s and 1980s attempts were made to build
database systems with integrated hardware and software.
The underlying philosophy was that such integration
would provide higher performance at lower cost. Examples were IBM System/38, the early offering of Teradata,
and the Britton Lee, Inc. database machine.
Another approach to hardware support for database management was ICL's CAFS accelerator, a hardware disk
controller with programmable search capabilities. In
the long term, these efforts were generally unsuccessful
because specialized database machines could not keep
pace with the rapid development and progress of generalpurpose computers. Thus most database systems nowadays are software systems running on general-purpose
hardware, using general-purpose computer data storage.
However this idea is still pursued for certain applications
by some companies like Netezza and Oracle (Exadata).
space allocation.”[19] dBASE was one of the top selling
software titles in the 1980s and early 1990s.
Late 1970s, SQL DBMS
1990s, object-oriented
IBM started working on a prototype system loosely based
on Codd’s concepts as System R in the early 1970s. The
first version was ready in 1974/5, and work then started
on multi-table systems in which the data could be split so
that all of the data for a record (some of which is optional)
did not have to be stored in a single large “chunk”. Subsequent multi-user versions were tested by customers in
1978 and 1979, by which time a standardized query language – SQL – had been added. Codd’s ideas were establishing themselves as both workable and superior to CODASYL, pushing IBM to develop a true production version of System R, known as SQL/DS, and, later, Database
2 (DB2).
Larry Ellison's Oracle started from a different chain,
based on IBM’s papers on System R, and beat IBM to
market when the first version was released in 1978.
The 1990s, along with a rise in object-oriented programming, saw a growth in how data in various databases were
handled. Programmers and designers began to treat the
data in their databases as objects. That is to say that if a
person’s data were in a database, that person’s attributes,
such as their address, phone number, and age, were now
considered to belong to that person instead of being extraneous data. This allows for relations between data to
be relations to objects and their attributes and not to individual fields.[20] The term "object-relational impedance
mismatch" described the inconvenience of translating between programmed objects and database tables. Object
databases and object-relational databases attempt to solve
this problem by providing an object-oriented language
(sometimes as extensions to SQL) that programmers can
use as alternative to purely relational SQL. On the programming side, libraries known as object-relational mappings (ORMs) attempt to solve the same problem.
Stonebraker went on to apply the lessons from INGRES
to develop a new database, Postgres, which is now known
as PostgreSQL. PostgreSQL is often used for global mission critical applications (the .org and .info domain name
registries use it as their primary data store, as do many 2000s, NoSQL and NewSQL
large companies and financial institutions).
In Sweden, Codd’s paper was also read and Mimer SQL Main articles: NoSQL and NewSQL
was developed from the mid-1970s at Uppsala University. In 1984, this project was consolidated into an independent enterprise. In the early 1980s, Mimer introduced
transaction handling for high robustness in applications,
an idea that was subsequently implemented on most other
Another data model, the entity–relationship model,
emerged in 1976 and gained popularity for database design as it emphasized a more familiar description than
the earlier relational model. Later on, entity–relationship
constructs were retrofitted as a data modeling construct
for the relational model, and the difference between the
two have become irrelevant.
1980s, on the desktop
The 1980s ushered in the age of desktop computing. The
new computers empowered their users with spreadsheets
like Lotus 1-2-3 and database software like dBASE. The
dBASE product was lightweight and easy for any computer user to understand out of the box. C. Wayne
Ratliff the creator of dBASE stated: “dBASE was different from programs like BASIC, C, FORTRAN, and
COBOL in that a lot of the dirty work had already been
done. The data manipulation is done by dBASE instead
of by the user, so the user can concentrate on what he
is doing, rather than having to mess with the dirty details of opening, reading, and closing files, and managing
The next generation of post-relational databases in the
2000s became known as NoSQL databases, including fast
key-value stores and document-oriented databases.
XML databases are a type of structured documentoriented database that allows querying based on XML
document attributes. XML databases are mostly used
in enterprise database management, where XML is being used as the machine-to-machine data interoperability standard. XML database management systems
include commercial software MarkLogic and Oracle
Berkeley DB XML, and a free use software Clusterpoint
Distributed XML/JSON Database. All are enterprise
software database platforms and support industry standard ACID-compliant transaction processing with strong
database consistency characteristics and high level of
database security.[21][22][23]
NoSQL databases are often very fast, do not require fixed table schemas, avoid join operations by
storing denormalized data, and are designed to scale
horizontally. The most popular NoSQL systems include MongoDB, Couchbase, Riak, Memcached, Redis,
CouchDB, Hazelcast, Apache Cassandra and HBase,[24]
which are all open-source software products.
In recent years there was a high demand for massively
distributed databases with high partition tolerance but
according to the CAP theorem it is impossible for a
distributed system to simultaneously provide consistency,
availability and partition tolerance guarantees. A distributed system can satisfy any two of these guarantees
at the same time, but not all three. For that reason many
NoSQL databases are using what is called eventual consistency to provide both availability and partition tolerance guarantees with a reduced level of data consistency.
NewSQL is a class of modern relational databases
that aims to provide the same scalable performance
of NoSQL systems for online transaction processing
(read-write) workloads while still using SQL and maintaining the ACID guarantees of a traditional database
system. Such databases include ScaleBase, Clustrix,
EnterpriseDB, MemSQL, NuoDB[25] and VoltDB.
Database technology has been an active research topic
since the 1960s, both in academia and in the research
and development groups of companies (for example IBM
Research). Research activity includes theory and development of prototypes. Notable research topics have included models, the atomic transaction concept and related concurrency control techniques, query languages
and query optimization methods, RAID, and more.
The database research area has several dedicated
academic journals (for example, ACM Transactions on
Database Systems-TODS, Data and Knowledge Engineering-DKE) and annual conferences (e.g., ACM SIGMOD,
One way to classify databases involves the type of their
contents, for example: bibliographic, document-text, statistical, or multimedia objects. Another way is by their
application area, for example: accounting, music compositions, movies, banking, manufacturing, or insurance.
A third way is by some technical aspect, such as the
database structure or interface type. This section lists a
few of the adjectives used to characterize different kinds
of databases.
• An in-memory database is a database that primarily
resides in main memory, but is typically backed-up
by non-volatile computer data storage. Main memory databases are faster than disk databases, and so
are often used where response time is critical, such
as in telecommunications network equipment.[26]
SAP HANA platform is a very hot topic for inmemory database. By May 2012, HANA was
able to run on servers with 100TB main memory
powered by IBM. The co founder of the company
claimed that the system was big enough to run the 8
largest SAP customers.
• An active database includes an event-driven archi-
tecture which can respond to conditions both inside
and outside the database. Possible uses include security monitoring, alerting, statistics gathering and authorization. Many databases provide active database
features in the form of database triggers.
• A cloud database relies on cloud technology. Both
the database and most of its DBMS reside remotely,
“in the cloud”, while its applications are both developed by programmers and later maintained and
utilized by (application’s) end-users through a web
browser and Open APIs.
• Data warehouses archive data from operational
databases and often from external sources such as
market research firms. The warehouse becomes the
central source of data for use by managers and other
end-users who may not have access to operational
data. For example, sales data might be aggregated
to weekly totals and converted from internal product codes to use UPCs so that they can be compared with ACNielsen data. Some basic and essential components of data warehousing include extracting, analyzing, and mining data, transforming,
loading and managing data so as to make them available for further use.
• A deductive database combines logic programming
with a relational database, for example by using the
Datalog language.
• A distributed database is one in which both the data
and the DBMS span multiple computers.
• A document-oriented database is designed for storing, retrieving, and managing document-oriented,
or semi structured data, information. Documentoriented databases are one of the main categories of
NoSQL databases.
• An embedded database system is a DBMS which is
tightly integrated with an application software that
requires access to stored data in such a way that the
DBMS is hidden from the application’s end-users
and requires little or no ongoing maintenance.[27]
• End-user databases consist of data developed by
individual end-users. Examples of these are collections of documents, spreadsheets, presentations,
multimedia, and other files. Several products exist to support such databases. Some of them are
much simpler than full-fledged DBMSs, with more
elementary DBMS functionality.
• A federated database system comprises several distinct databases, each with its own DBMS. It is handled as a single database by a federated database
management system (FDBMS), which transparently
integrates multiple autonomous DBMSs, possibly of
different types (in which case it would also be a
heterogeneous database system), and provides them
with an integrated conceptual view.
• Sometimes the term multi-database is used as a synonym to federated database, though it may refer
to a less integrated (e.g., without an FDBMS and
a managed integrated schema) group of databases
that cooperate in a single application. In this
case typically middleware is used for distribution,
which typically includes an atomic commit protocol
(ACP), e.g., the two-phase commit protocol, to allow distributed (global) transactions across the participating databases.
• A graph database is a kind of NoSQL database that
uses graph structures with nodes, edges, and properties to represent and store information. General
graph databases that can store any graph are distinct
from specialized graph databases such as triplestores
and network databases.
• An array DBMS is a kind of NoSQL DBMS that
allows to model, store, and retrieve (usually large)
multi-dimensional arrays such as satellite images
and climate simulation output.
• In a hypertext or hypermedia database, any word or
a piece of text representing an object, e.g., another
piece of text, an article, a picture, or a film, can be
hyperlinked to that object. Hypertext databases are
particularly useful for organizing large amounts of
disparate information. For example, they are useful
for organizing online encyclopedias, where users can
conveniently jump around the text. The World Wide
Web is thus a large distributed hypertext database.
• A knowledge base (abbreviated KB, kb or Δ
is a special kind of database for knowledge management, providing the means for the computerized
collection, organization, and retrieval of knowledge.
Also a collection of data representing problems with
their solutions and related experiences.
• A mobile database can be carried on or synchronized
from a mobile computing device.
The major parallel DBMS architectures which are induced by the underlying hardware architecture are:
• Shared memory architecture,
where multiple
processors share the main
memory space, as well as
other data storage.
• Shared disk architecture,
where each processing unit
(typically consisting of multiple processors) has its own
main memory, but all units
share the other storage.
• Shared nothing architecture, where each processing
unit has its own main memory
and other storage.
• Probabilistic databases employ fuzzy logic to draw
inferences from imprecise data.
• Real-time databases process transactions fast
enough for the result to come back and be acted on
right away.
• A spatial database can store the data with multidimensional features. The queries on such data include location based queries, like “Where is the closest hotel in my area?".
• A temporal database has built-in time aspects, for
example a temporal data model and a temporal version of SQL. More specifically the temporal aspects
usually include valid-time and transaction-time.
• A terminology-oriented database builds upon an
object-oriented database, often customized for a
specific field.
• An unstructured data database is intended to store
in a manageable and protected way diverse objects
that do not fit naturally and conveniently in common databases. It may include email messages,
documents, journals, multimedia objects, etc. The
name may be misleading since some objects can be
highly structured. However, the entire possible object collection does not fit into a predefined structured framework. Most established DBMSs now
support unstructured data in various ways, and new
dedicated DBMSs are emerging.
• Operational databases store detailed data about
the operations of an organization. They typically
process relatively high volumes of updates using
transactions. Examples include customer databases
that record contact, credit, and demographic information about a business’ customers, personnel
databases that hold information such as salary, benefits, skills data about employees, enterprise resource planning systems that record details about
product components, parts inventory, and finan1.1.7 Design and modeling
cial databases that keep track of the organization’s
money, accounting and financial dealings.
Main article: Database design
• A parallel database seeks to improve performance
through parallelization for tasks such as loading data, The first task of a database designer is to produce a
building indexes and evaluating queries.
conceptual data model that reflects the structure of the
information to be held in the database. A common approach to this is to develop an entity-relationship model,
often with the aid of drawing tools. Another popular approach is the Unified Modeling Language. A successful
data model will accurately reflect the possible state of the
external world being modeled: for example, if people can
have more than one phone number, it will allow this information to be captured. Designing a good conceptual
data model requires a good understanding of the application domain; it typically involves asking deep questions
about the things of interest to an organisation, like “can
a customer also be a supplier?", or “if a product is sold
with two different forms of packaging, are those the same
product or different products?", or “if a plane flies from
New York to Dubai via Frankfurt, is that one flight or two
(or maybe even three)?". The answers to these questions
establish definitions of the terminology used for entities
(customers, products, flights, flight segments) and their
relationships and attributes.
requirements, and requires a good knowledge of the expected workload and access patterns, and a deep understanding of the features offered by the chosen DBMS.
Another aspect of physical database design is security. It
involves both defining access control to database objects
as well as defining security levels and methods for the data
Main article: Database model
A database model is a type of data model that deter-
Producing the conceptual data model sometimes involves
input from business processes, or the analysis of workflow
in the organization. This can help to establish what information is needed in the database, and what can be left
out. For example, it can help when deciding whether the
database needs to hold historic data as well as current
Having produced a conceptual data model that users are
happy with, the next stage is to translate this into a schema
that implements the relevant data structures within the
database. This process is often called logical database
design, and the output is a logical data model expressed
in the form of a schema. Whereas the conceptual data
model is (in theory at least) independent of the choice
of database technology, the logical data model will be
expressed in terms of a particular database model supported by the chosen DBMS. (The terms data model and
database model are often used interchangeably, but in
this article we use data model for the design of a specific
database, and database model for the modelling notation
used to express that design.)
The most popular database model for general-purpose
databases is the relational model, or more precisely, the
relational model as represented by the SQL language.
The process of creating a logical database design using this model uses a methodical approach known as
normalization. The goal of normalization is to ensure
that each elementary “fact” is only recorded in one place,
so that insertions, updates, and deletions automatically
maintain consistency.
Collage of five types of database models
mines the logical structure of a database and fundamentally determines in which manner data can be stored, organized, and manipulated. The most popular example of
a database model is the relational model (or the SQL approximation of relational), which uses a table-based format.
Common logical data models for databases include:
• Navigational databases
• Hierarchical database model
• Network model
• Graph database
• Relational model
• Entity–relationship model
• Enhanced entity–relationship model
• Object model
• Document model
The final stage of database design is to make the decisions
• Entity–attribute–value model
that affect performance, scalability, recovery, security,
and the like. This is often called physical database design.
• Star schema
A key goal during this stage is data independence, meaning that the decisions made for performance optimizaAn object-relational database combines the two related
tion purposes should be invisible to end-users and applistructures.
cations. Physical design is driven mainly by performance
Physical data models include:
• Inverted index
• Flat file
Other models include:
• Associative model
• Multidimensional model
• Array model
• Multivalue model
• The internal level (or physical level) is the internal organization of data inside a DBMS. It is concerned with cost, performance, scalability and other
operational matters. It deals with storage layout of
the data, using storage structures such as indexes to
enhance performance. Occasionally it stores data
of individual views (materialized views), computed
from generic data, if performance justification exists for such redundancy. It balances all the external
views’ performance requirements, possibly conflicting, in an attempt to optimize overall performance
across all activities.
Specialized models are optimized for particular types of While there is typically only one conceptual (or logical)
and physical (or internal) view of the data, there can be
any number of different external views. This allows users
to see database information in a more business-related
• XML database
way rather than from a technical, processing viewpoint.
For example, a financial department of a company needs
• Semantic model
the payment details of all employees as part of the com• Content store
pany’s expenses, but does not need details about employees that are the interest of the human resources depart• Event store
ment. Thus different departments need different views
of the company’s database.
• Time series model
The three-level database architecture relates to the concept of data independence which was one of the major
External, conceptual, and internal views
initial driving forces of the relational model. The idea
is that changes made at a certain level do not affect the
view at a higher level. For example, changes in the internal level do not affect application programs written using
conceptual level interfaces, which reduces the impact of
making physical changes to improve performance.
The conceptual view provides a level of indirection between internal and external. On one hand it provides a
common view of the database, independent of different
external view structures, and on the other hand it abstracts
away details of how the data is stored or managed (internal level). In principle every level, and even every exterInternal Schema
External Schema
nal view, can be presented by a different data model. In
-User View-Computer Viewpractice usually a given DBMS uses the same data model
for both the external and the conceptual levels (e.g., relational model). The internal level, which is hidden inside
Traditional view of data
the DBMS and depends on its implementation, requires
a different level of detail and uses its own types of data
A database management system provides three views of
structure types.
the database data:
Separating the external, conceptual and internal levels was
major feature of the relational database model imple• The external level defines how each group of endmentations
that dominate 21st century databases.[31]
users sees the organization of data in the database.
A single database can have any number of views at
the external level.
1.1.8 Languages
• The conceptual level unifies the various external
views into a compatible global view.[31] It provides Database languages are special-purpose languages, which
the synthesis of all the external views. It is out of do one or more of the following:
the scope of the various database end-users, and is
• Data definition language – defines data types and the
rather of interest to database application developers
relationships among them
and database administrators.
• Data manipulation language – performs tasks such Storage
as inserting, updating, or deleting data occurrences
Main articles: Computer data storage and Database
• Query language – allows searching for information engine
and computing derived information
Database languages are specific to a particular data Database storage is the container of the physical materialization of a database. It comprises the internal (physmodel. Notable examples include:
ical) level in the database architecture. It also contains
all the information needed (e.g., metadata, “data about
• SQL combines the roles of data definition, data mathe data”, and internal data structures) to reconstruct the
nipulation, and query in a single language. It was
conceptual level and external level from the internal level
one of the first commercial languages for the relawhen needed. Putting data into permanent storage is
tional model, although it departs in some respects
generally the responsibility of the database engine a.k.a.
from the relational model as described by Codd (for
“storage engine”. Though typically accessed by a DBMS
example, the rows and columns of a table can be orthrough the underlying operating system (and often utidered). SQL became a standard of the American
lizing the operating systems’ file systems as intermediates
National Standards Institute (ANSI) in 1986, and of
for storage layout), storage properties and configuration
the International Organization for Standardization
setting are extremely important for the efficient opera(ISO) in 1987. The standards have been regularly
tion of the DBMS, and thus are closely maintained by
enhanced since and is supported (with varying dedatabase administrators. A DBMS, while in operation,
grees of conformance) by all mainstream commeralways has its database residing in several types of storage
cial relational DBMSs.[32][33]
(e.g., memory and external storage). The database data
and the additional needed information, possibly in very
• OQL is an object model language standard (from the large amounts, are coded into bits. Data typically reside
Object Data Management Group). It has influenced in the storage in structures that look completely different
the design of some of the newer query languages like from the way the data look in the conceptual and external
levels, but in ways that attempt to optimize (the best pos• XQuery is a standard XML query language im- sible) these levels’ reconstruction when needed by users
plemented by XML database systems such as and programs, as well as for computing additional types
MarkLogic and eXist, by relational databases with of needed information from the data (e.g., when querying
XML capability such as Oracle and DB2, and also the database).
by in-memory XML processors such as Saxon.
• SQL/XML combines XQuery with SQL.[34]
Some DBMSs support specifying which character encoding was used to store data, so multiple encodings can be
used in the same database.
Various low-level database storage structures are used by
the storage engine to serialize the data model so it can
be written to the medium of choice. Techniques such as
• DBMS-specific Configuration and storage engine
indexing may be used to improve performance. Convenmanagement
tional storage is row-oriented, but there are also column• Computations to modify query results, like count- oriented and correlation databases.
ing, summing, averaging, sorting, grouping, and
Materialized views Main article: Materialized view
• Constraint enforcement (e.g. in an automotive
database, only allowing one engine type per car)
Often storage redundancy is employed to increase per• Application programming interface version of the formance. A common example is storing materialized
query language, for programmer convenience
views, which consist of frequently needed external views
or query results. Storing such views saves the expensive
computing of them each time they are needed. The down1.1.9 Performance, security, and availabil- sides of materialized views are the overhead incurred
when updating them to keep them synchronized with their
original updated database data, and the cost of storage reBecause of the critical importance of database technology dundancy.
to the smooth running of an enterprise, database systems
include complex mechanisms to deliver the required performance, security, and availability, and allow database Replication Main article: Database replication
administrators to control the use of these features.
A database language may also incorporate features like:
Occasionally a database employs storage redundancy by
database objects replication (with one or more copies) to
increase data availability (both to improve performance
of simultaneous multiple end-user accesses to a same
database object, and to provide resiliency in a case of partial failure of a distributed database). Updates of a replicated object need to be synchronized across the object
copies. In many cases the entire database is replicated.
changes. Sometimes application-level code is used to
record changes rather than leaving this to the database.
Monitoring can be set up to attempt to detect security
Transactions and concurrency
Further information: Concurrency control
Main article: Database security
Database security deals with all various aspects of protecting the database content, its owners, and its users.
It ranges from protection from intentional unauthorized
database uses to unintentional database accesses by unauthorized entities (e.g., a person or a computer program).
Database access control deals with controlling who (a
person or a certain computer program) is allowed to access what information in the database. The information may comprise specific database objects (e.g., record
types, specific records, data structures), certain computations over certain objects (e.g., query types, or specific
queries), or utilizing specific access paths to the former
(e.g., using specific indexes or other data structures to access information). Database access controls are set by
special authorized (by the database owner) personnel that
uses dedicated protected security DBMS interfaces.
This may be managed directly on an individual basis, or
by the assignment of individuals and privileges to groups,
or (in the most elaborate models) through the assignment
of individuals and groups to roles which are then granted
entitlements. Data security prevents unauthorized users
from viewing or updating the database. Using passwords,
users are allowed access to the entire database or subsets of it called “subschemas”. For example, an employee
database can contain all the data about an individual employee, but one group of users may be authorized to view
only payroll data, while others are allowed access to only
work history and medical data. If the DBMS provides
a way to interactively enter and update the database, as
well as interrogate it, this capability allows for managing
personal databases.
Data security in general deals with protecting specific
chunks of data, both physically (i.e., from corruption, or
destruction, or removal; e.g., see physical security), or the
interpretation of them, or parts of them to meaningful information (e.g., by looking at the strings of bits that they
comprise, concluding specific valid credit-card numbers;
e.g., see data encryption).
Database transactions can be used to introduce some level
of fault tolerance and data integrity after recovery from a
crash. A database transaction is a unit of work, typically
encapsulating a number of operations over a database
(e.g., reading a database object, writing, acquiring lock,
etc.), an abstraction supported in database and also other
systems. Each transaction has well defined boundaries
in terms of which program/code executions are included
in that transaction (determined by the transaction’s programmer via special transaction commands).
The acronym ACID describes some ideal properties of a
database transaction: Atomicity, Consistency, Isolation,
and Durability.
See also section Database migration in article
Data migration
A database built with one DBMS is not portable to another DBMS (i.e., the other DBMS cannot run it). However, in some situations it is desirable to move, migrate
a database from one DBMS to another. The reasons are
primarily economical (different DBMSs may have different total costs of ownership or TCOs), functional, and operational (different DBMSs may have different capabilities). The migration involves the database’s transformation from one DBMS type to another. The transformation
should maintain (if possible) the database related application (i.e., all related application programs) intact. Thus,
the database’s conceptual and external architectural levels should be maintained in the transformation. It may be
desired that also some aspects of the architecture internal
level are maintained. A complex or large database migration may be a complicated and costly (one-time) project
by itself, which should be factored into the decision to migrate. This in spite of the fact that tools may exist to help
migration between specific DBMSs. Typically a DBMS
vendor provides tools to help importing databases from
other popular DBMSs.
Change and access logging records who accessed which Building, maintaining, and tuning
attributes, what was changed, and when it was changed.
Logging services allow for a forensic database audit Main article: Database tuning
later by keeping a record of access occurrences and
After designing a database for an application, the next Other
stage is building the database. Typically an appropriate general-purpose DBMS can be selected to be utilized Other DBMS features might include:
for this purpose. A DBMS provides the needed user interfaces to be utilized by database administrators to de• Database logs
fine the needed application’s data structures within the
• Graphics component for producing graphs and
DBMS’s respective data model. Other user interfaces are
charts, especially in a data warehouse system
used to select needed DBMS parameters (like security
related, storage allocation parameters, etc.).
• Query optimizer – Performs query optimization on
every query to choose for it the most efficient query
When the database is ready (all its data structures and
plan (a partial order (tree) of operations) to be exeother needed components are defined) it is typically popcuted to compute the query result. May be specific
ulated with initial application’s data (database initializato a particular storage engine.
tion, which is typically a distinct project; in many cases
using specialized DBMS interfaces that support bulk in• Tools or hooks for database design, application
sertion) before making it operational. In some cases the
programming, application program maintenance,
database becomes operational while empty of application
database performance analysis and monitoring,
data, and data is accumulated during its operation.
database configuration monitoring, DBMS hardAfter the database is created, initialised and populated
ware configuration (a DBMS and related database
it needs to be maintained. Various database paramemay span computers, networks, and storage units)
ters may need changing and the database may need to be
and related database mapping (especially for a distuned (tuning) for better performance; application’s data
tributed DBMS), storage allocation and database
structures may be changed or added, new related applicalayout monitoring, storage migration, etc.
tion programs may be written to add to the application’s
functionality, etc.
1.1.10 See also
Backup and restore
Main article: Outline of databases
Main article: Backup
• Comparison of database tools
Sometimes it is desired to bring a database back to a
previous state (for many reasons, e.g., cases when the
database is found corrupted due to a software error, or if
it has been updated with erroneous data). To achieve this
a backup operation is done occasionally or continuously,
where each desired database state (i.e., the values of its
data and their embedding in database’s data structures) is
kept within dedicated backup files (many techniques exist to do this effectively). When this state is needed, i.e.,
when it is decided by a database administrator to bring the
database back to this state (e.g., by specifying this state
by a desired point in time when the database was in this
state), these files are utilized to restore that state.
• Comparison of object database management systems
• Comparison of object-relational database management systems
• Comparison of relational database management systems
• Data hierarchy
• Data bank
• Data store
• Database theory
Static Analysis
• Database testing
• Database-centric architecture
Static analysis techniques for software verification can
be applied also in the scenario of query languages. In
• Question-focused dataset
particular, the *Abstract interpretation framework has
been extended to the field of query languages for relational databases as a way to support sound approximation 1.1.11 References
techniques.[35] The semantics of query languages can be
tuned according to suitable abstractions of the concrete [1] “Database - Definition of database by Merriam-Webster”.
domain of data. The abstraction of relational database
system has many interesting applications, in particular, [2] Jeffrey Ullman 1997: First course in database systems,
for security purposes, such as fine grained access control,
Prentice–Hall Inc., Simon & Schuster, Page 1, ISBN 013-861337-0.
watermarking, etc.
[3] “Update - Definition of update by Merriam-Webster”.
[4] “Retrieval - Definition of retrieval by Merriam-Webster”.
[5] “Administration - Definition of administration by
[6] Tsitchizris, D. C. and F. H. Lochovsky (1982). Data Models. Englewood-Cliffs, Prentice–Hall.
[7] Beynon-Davies P. (2004). Database Systems 3rd Edition.
Palgrave, Basingstoke, UK. ISBN 1-4039-1601-2
[8] Raul F. Chong, Michael Dang, Dwaine R. Snow, Xiaomei
Wang (3 July 2008). “Introduction to DB2”. Retrieved
17 March 2013.. This article quotes a development time
of 5 years involving 750 people for DB2 release 9 alone
[9] C. W. Bachmann (November 1973), “The Programmer as
Navigator” (PDF), CACM (Turing Award Lecture 1973)
[10] “TOPDB Top Database index”.
[11] “database, n”. OED Online. Oxford University Press.
June 2013. Retrieved July 12, 2013.
[12] IBM Corporation. “IBM Information Management System (IMS) 13 Transaction and Database Servers delivers
high performance and low total cost of ownership”. Retrieved Feb 20, 2014.
[13] Codd, E.F. (1970).“A Relational Model of Data for Large
Shared Data Banks”. In: Communications of the ACM 13
(6): 377–387.
[20] Development of an object-oriented DBMS; Portland, Oregon, United States; Pages: 472 – 482; 1986; ISBN 089791-204-7
[21] “Oracle Berkeley DB XML” (PDF). Retrieved 10 March
[22] “ACID Transactions, MarkLogic”. Retrieved 10 March
[23] “Clusterpoint Database at a Glance”. Retrieved 10 March
[24] “DB-Engines Ranking”. January 2013. Retrieved 22 January 2013.
[25] Proctor, Seth (2013). “Exploring the Architecture of the
NuoDB Database, Part 1”. Retrieved 2013-07-12.
[26] “TeleCommunication Systems Signs up as a Reseller of
TimesTen; Mobile Operators and Carriers Gain RealTime Platform for Location-Based Services”. Business
Wire. 2002-06-24.
[27] Graves, Steve. “COTS Databases For Embedded Systems”, Embedded Computing Design magazine, January
2007. Retrieved on August 13, 2008.
[28] Argumentation in Artificial Intelligence by Iyad Rahwan,
Guillermo R. Simari
[29] “OWL DL Semantics”. Retrieved 10 December 2010.
[30] (1993) Integration Definition for Information
Modeling (IDEFIX). 21 December 1993.
[31] Date 1990, pp. 31–32
[14] William Hershey and Carol Easthope, “A set theoretic data structure and retrieval language”, Spring Joint
Computer Conference, May 1972 in ACM SIGIR Forum, Volume 7, Issue 4 (December 1972), pp. 45–55,
[32] Chapple, Mike. “SQL Fundamentals”. Retrieved 2009-01-28.
[15] Ken North, “Sets, Data Models and Data Independence”,
Dr. Dobb’s, 10 March 2010
[34] Wagner, Michael (2010), “1. Auflage”, SQL/XML:2006
– Evaluierung der Standardkonformität ausgewählter
Datenbanksysteme, Diplomica Verlag, ISBN 3-83669609-6
[16] Description of a set-theoretic data structure, D. L. Childs,
1968, Technical Report 3 of the CONCOMP (Research
in Conversational Use of Computers) Project, University
of Michigan, Ann Arbor, Michigan, USA
[17] Feasibility of a Set-Theoretic Data Structure : A General
Structure Based on a Reconstituted Definition of Relation,
D. L. Childs, 1968, Technical Report 6 of the CONCOMP (Research in Conversational Use of Computers)
Project, University of Michigan, Ann Arbor, Michigan,
[18] MICRO Information Management System (Version 5.0)
Reference Manual, M.A. Kahn, D.L. Rumelhart, and B.L.
Bronson, October 1977, Institute of Labor and Industrial
Relations (ILIR), University of Michigan and Wayne State
[19] Interview with Wayne Ratliff. The FoxPro History. Retrieved on 2013-07-12.
[33] “Structured Query Language (SQL)". International Business Machines. October 27, 2006. Retrieved 2007-0610.
[35] R.Halder and A.Cortesi, Abstract Interpretation of
Database Query Languages.
123-−157, Elsevier Ed. (ISSN 1477-8424)
1.1.12 Further reading
• Ling Liu and Tamer M. Özsu (Eds.) (2009).
"Encyclopedia of Database Systems, 4100 p. 60 illus. ISBN 978-0-387-49616-0.
• Beynon-Davies, P. (2004). Database Systems. 3rd
Edition. Palgrave, Houndmills, Basingstoke.
• Connolly, Thomas and Carolyn Begg. Database Systems. New York: Harlow, 2002.
• Date, C. J. (2003). An Introduction to Database Sys- Instead, the tools help to preserve the meaning of the data
tems, Fifth Edition. Addison Wesley. ISBN 0-201- or to reorganize existing data to meet new requirements.
Since meaning of the data often cannot be encoded, the
configuration of the tools usually needs manual interven• Gray, J. and Reuter, A. Transaction Processing: tion.
Concepts and Techniques, 1st edition, Morgan Kaufmann Publishers, 1992.
• Kroenke, David M. and David J. Auer. Database
Concepts. 3rd ed. New York: Prentice, 2007.
1.2.1 Risks and Benefits
Schema migration allows to fix mistakes and adapt the
• Raghu Ramakrishnan and Johannes Gehrke, data as requirements change. They are an essential part of
software evolution, especially in agile environments (see
Database Management Systems
• Abraham Silberschatz, Henry F. Korth, S. SudarApplying a schema migration to a production database is
shan, Database System Concepts
always a risk. Development and test databases tend to be
• Discussion on database systems,
smaller and cleaner. The data in them is better understood
or, if everything else fails, the amount of data is small
• Lightstone, S.; Teorey, T.; Nadeau, T. (2007). Physenough for a human to process. Production databases are
ical Database Design: the database professional’s
usually huge, old and full of surprises. The surprises can
guide to exploiting indexes, views, storage, and more.
come from many sources:
Morgan Kaufmann Press. ISBN 0-12-369389-6.
• Teorey, T.; Lightstone, S. and Nadeau, T. Database
Modeling & Design: Logical Design, 4th edition,
Morgan Kaufmann Press, 2005. ISBN 0-12685352-5
External links
• Database at DMOZ
• DB File extension – informations about files with
DB extension
1.2 Schema migration
Not to be confused with Data migration.
In software engineering, schema migration (also
database migration, database change management[1][2] ) refers to the management of incremental,
reversible changes to relational database schemas. A
schema migration is performed on a database whenever
it is necessary to update or revert that database’s schema
to some newer or older version.
• Corrupt data that was written by old versions of the
software and not cleaned properly
• Implied dependencies in the data which no one
knows about anymore
• People directly changing the database without using
the designated tools
• Bugs in the schema migration tools
• Mistakes in assumptions how data should be migrated
For these reasons, the migration process needs a high
level of discipline, thorough testing and a sound backup
1.2.2 Schema migration in agile software
Migrations are performed programmatically by using a
schema migration tool. When invoked with a specified
desired schema version, the tool automates the successive application or reversal of an appropriate sequence of
schema changes until it is brought to the desired state.
When developing software applications backed by a
database, developers typically develop the application
source code in tandem with an evolving database
schema. The code typically has rigid expectations of
what columns, tables and constraints are present in the
database schema whenever it needs to interact with one,
so only the version of database schema against which the
code was developed is considered fully compatible with
that version of source code.
Most schema migration tools aim to minimise the impact
of schema changes on any existing data in the database.
Despite this, preservation of data in general is not guaranteed because schema changes such as the deletion of a
database column can destroy data (i.e. all values stored
under that column for all rows in that table are deleted).
In software testing, while developers may mock the presence of a compatible database system for unit testing,
any level of testing higher than this (e.g. integration testing or system testing) it is common for developers to test
their application against a local or remote test database
schematically compatible with the version of source code
under test. In advanced applications, the migration itself 1.2.3 Available Tools
can be subject to migration testing.
• Flyway - database migration tool (for Windows,
With schema migration technology, data models no
OSX, Linux, Android and the JVM) where migralonger need to be fully designed up-front, and is more
tions are written in SQL or Java
capable of being adapted with changing project requirements throughout the software development lifecycle.
• LiquiBase - cross platform tool where migrations are
written in XML, YAML, JSON or SQL.
Relation to revision control systems
Teams of software developers usually use version control systems to manage and collaborate on changes made
to versions of source code. Different developers can develop on divergent, relatively older or newer branches of
the same source code to make changes and additions during development.
Supposing that the software under development interacts
with a database, every version of the source code can be
associated with at least one database schema with which
it is compatible.
Under good software testing practise, schema migrations
can be performed on test databases to ensure that their
schema is compatible to the source code. To streamline
this process, a schema migration tool is usually invoked
as a part of an automated software build as a prerequisite
of the automated testing phase.
Schema migration tools can be said to solve versioning
problems for database schemas just as version control systems solve versioning problems for source code. In practice, many schema migration tools actually rely on a textual representation of schema changes (such as files containing SQL statements) such that the version history of
schema changes can effectively be stored alongside program source code within VCS. This approach ensures
that the information necessary to recover a compatible
database schema for a particular code branch is recoverable from the source tree itself. Another benefit of this
approach is the handling of concurrent conflicting schema
changes; developers may simply use their usual text-based
conflict resolution tools to reconcile differences.
• Datical - Enterprise commercial version of
• Redgate SQL Compare - a schema comparison and
deployment tool for SQL Server and Oracle.
• ReadyRoll - a migrations-based Visual Studio extension for SQL Server development and deployment.
• Active Record (Migrations) - schema migration tool
for Ruby on Rails projects based on Active Record.
• Ruckusing-migrations - schema migration tool for
PHP projects.
• Phinx - another framework-independent PHP migration tool.
• MyBatis Migrations - seeks to be the best migration
tool of its kind.
• Ragtime - a SQL database schema migration library
written in Clojure
• Lobos - a SQL database schema manipulation and
migration library written in Clojure.
• Alembic - a lightweight database migration tool for
usage with the SQLAlchemy Database Toolkit for
• RoundhousE - a SQL database versioning and
change management tool written in C#.
• XMigra - a SQL database evolution management
tool written in Ruby that generates scripts without
communicating with the database.
Relation to schema evolution
• DBmaestro - a database version control and schema
migration solution for SQL Server and Oracle.
Schema migration tooling could be seen as a facility to
track the history of an evolving schema.
• DB Change Manager - Commercial Change Management Software by Embarcadero.
• Sqitch - Sqitch by Theory.
Developers no longer need to remove the entire test 1.2.4 References
database in order to create a new test database from
[1] Liquibase Database Refactorscratch (e.g. using schema creation scripts from DDL
generation tools). Further, if generation of test data costs
a lot of time, developers can avoid regenerating test data [2] Flyway: The agile database migration framework for Java
for small, non-destructive changes to the schema.
1.3 Star schema
Dimension tables
In computing, the Star Schema is the simplest style of
data mart schema. The star schema consists of one or
more fact tables referencing any number of dimension
tables. The star schema is an important special case of
the snowflake schema, and is more effective for handling
simpler queries.[1]
Dimension tables usually have a relatively small number
of records compared to fact tables, but each record may
have a very large number of attributes to describe the fact
data. Dimensions can define a wide variety of characteristics, but some of the most common attributes defined
by dimension tables include:
The star schema gets its name from the physical model’s[2]
resemblance to a star shape with a fact table at its center
and the dimension tables surrounding it representing the
star’s points.
• Time dimension tables describe time at the lowest level of time granularity for which events are
recorded in the star schema
• Product dimension tables describe products
• Geography dimension tables describe location data,
such as country, state, or city
• Employee dimension tables describe employees,
The star schema separates business process data into
such as sales people
facts, which hold the measurable, quantitative data about
• Range dimension tables describe ranges of time,
a business, and dimensions which are descriptive atdollar values, or other measurable quantities to simtributes related to fact data. Examples of fact data include
plify reporting
sales price, sale quantity, and time, distance, speed, and
weight measurements. Related dimension attribute examples include product models, product colors, product Dimension tables are generally assigned a surrogate primary key, usually a single-column integer data type,
sizes, geographic locations, and salesperson names.
mapped to the combination of dimension attributes that
A star schema that has many dimensions is sometimes form the natural key.
called a centipede schema.[3] Having dimensions of only
a few attributes, while simpler to maintain, results in
queries with many table joins and makes the star schema 1.3.2 Benefits
less easy to use.
Star schemas are denormalized, meaning the normal
rules of normalization applied to transactional relational databases are relaxed during star schema deFact tables
sign and implementation. The benefits of star schema
Fact tables record measurements or metrics for a specific denormalization are:
event. Fact tables generally consist of numeric values, and
foreign keys to dimensional data where descriptive information is kept.[3] Fact tables are designed to a low level
of uniform detail (referred to as “granularity” or “grain”),
meaning facts can record events at a very atomic level.
This can result in the accumulation of a large number of
records in a fact table over time. Fact tables are defined
as one of three types:
• Transaction fact tables record facts about a specific
event (e.g., sales events)
• Snapshot fact tables record facts at a given point in
time (e.g., account details at month end)
• Accumulating snapshot tables record aggregate facts
at a given point in time (e.g., total month-to-date
sales for a product)
Fact tables are generally assigned a surrogate key to ensure each row can be uniquely identified.
• Simpler queries - star schema join logic is generally
simpler than the join logic required to retrieve data
from a highly normalized transactional schemas.
• Simplified business reporting logic - when compared
to highly normalized schemas, the star schema simplifies common business reporting logic, such as
period-over-period and as-of reporting.
• Query performance gains - star schemas can provide
performance enhancements for read-only reporting
applications when compared to highly normalized
• Fast aggregations - the simpler queries against a star
schema can result in improved performance for aggregation operations.
• Feeding cubes - star schemas are used by all OLAP
systems to build proprietary OLAP cubes efficiently; in fact, most major OLAP systems provide
a ROLAP mode of operation which can use a star
schema directly as a source without building a proprietary cube structure.
The main disadvantage of the star schema is that data integrity is not enforced as well as it is in a highly normalized database. One-off inserts and updates can result in
data anomalies which normalized schemas are designed
to avoid. Generally speaking, star schemas are loaded in
a highly controlled fashion via batch processing or nearreal time “trickle feeds”, to compensate for the lack of
protection afforded by normalization.
S.Country AS Countries,
SUM(F.Units_Sold) FROM Fact_Sales F INNER
JOIN Dim_Date D ON (F.Date_Id = D.Id) INNER
JOIN Dim_Store S ON (F.Store_Id = S.Id) INNER
JOIN Dim_Product P ON (F.Product_Id = P.Id)
WHERE D.Year = 1997 AND P.Product_Category =
'tv' GROUP BY P.Brand, S.Country
1.3.5 See also
Star schema is also not as flexible in terms of analytical
needs as a normalized data model. Normalized models al• Online analytical processing
low any kind of analytical queries to be executed as long
• Reverse star schema
as they follow the business logic defined in the model.
Star schemas tend to be more purpose-built for a partic• Snowflake schema
ular view of the data, thus not really allowing more complex analytics. Star schemas don't support many-to-many
• Fact constellation
relationships between business entities - at least not very
naturally. Typically these relationships are simplified in
star schema to conform to the simple dimensional model. 1.3.6 References
[1] “DWH Schemas”. 2009.
[2] C J Date, “An Introduction to Database Systems (Eighth
Edition)", p. 708
[3] Ralph Kimball and Margy Ross, The Data Warehouse
Toolkit: The Complete Guide to Dimensional Modeling
(Second Edition), p. 393
1.3.7 External links
• Designing the Star Schema Database by Craig Utley
• Stars: A Pattern Language for Query Optimized
Star schema used by example query.
Consider a database of sales, perhaps from a store chain,
classified by date, store and product. The image of the
schema to the right is a star schema version of the sample
schema provided in the snowflake schema article.
Fact_Sales is the fact table and there are three dimension
tables Dim_Date, Dim_Store and Dim_Product.
Each dimension table has a primary key on its Id column, relating to one of the columns (viewed as rows
in the example schema) of the Fact_Sales table’s threecolumn (compound) primary key (Date_Id, Store_Id,
Product_Id). The non-primary key Units_Sold column
of the fact table in this example represents a measure or
metric that can be used in calculations and analysis. The
non-primary key columns of the dimension tables represent additional attributes of the dimensions (such as the
Year of the Dim_Date dimension).
For example, the following query answers how many TV
sets have been sold, for each brand and country, in 1997:
• Fact constellation schema
• Data Warehouses, Schemas and Decision Support
Basics by Dan Power
Chapter 2
Not Only SQL
2.1 CAP
• Central Atlanta Progress
CAP may refer to:
• Chicago Area Project, a juvenile delinquency
• Christian Appalachian Project, a program to assist
disadvantaged persons in Kentucky and West Virginia
Science and medicine
• CaP, prostate cancer
• CAP (protein), cyclase-associated protein
• Christians Against Poverty, the UK charity
• Carrierless amplitude phase modulation
• Church Action on Poverty, UK national ecumenical
social justice charity established in 1982
• Catabolite activator protein, a regulatory protein for
mRNA transcription in prokaryotes that binds cyclic
• College of American Pathologists
• Cellulose acetate phthalate, a cellulose-based polymer
• Committee of Advertising Practice
• Committee for Another Policy (Comité voor een Andere Politiek / Comité pour une Autre Politique), a
Belgian political movement
• Community-acquired pneumonia
• Concerned Alumni of Princeton
• CAP computer, an experimental machine built in
Cambridge, UK
• Congress of Aboriginal Peoples, Canadian aboriginal organization
• CAP theorem, Consistency, Availability, Partitiontolerance theorem in computer science
• Camel Application Part, a protocol used in CAMEL
2.1.4 Companies
• Common Alerting Protocol, an XML based data
format for exchanging public warnings between different alerting technologies
• Companhia Aeronáutica Paulista, a 1940s Brazilian
aircraft manufacturer
• CAP Group (Computer Analysts and Programmers), a UK software company
• CAP S.A. (Compañía de Acero del Pacífico), a
Chilean mining and steel sector holding company
• Canadian Action Party
• Canadian Association of Physicists
• CAP Scientific, a British defence software company
• Center for Adoption Policy
• Center for American Progress, a left-of-centre think
• Constructions Aéronautiques Parisiennes, Apex
Aircraft training and aerobatic aircraft
Projects, programs, policies
• Common Agricultural Policy, the European Union’s
agricultural subsidy system
• Codice di Avviamento Postale, literally Postal Expedition Code, Italy’s postal code system
• Estadio CAP (Compañía de Acero del Pacífico), a
football stadium in Talcahuano, Chile
• Community Access Program, a government of
Canada initiative to provide access to the Internet
in remote areas
See also
• Capital Assistance Program
• CAP code (disambiguation)
• Community Action Program, Lyndon Johnson’s
anti-poverty programs
• Cap (disambiguation)
• Community Action Programme, United Kingdom
workfare scheme
• Combat air patrol
• Combined Action Program (AKA Combined Action Platoon), a United States Marine Corps Vietnam era special operation
• Civil Air Patrol, the official US Air Force Auxiliary
• Certified Automation Professional, certification
from the International Society of Automation
2.2 Eventual consistency
Eventual consistency is a consistency model used in
distributed computing to achieve high availability that informally guarantees that, if no new updates are made to
a given data item, eventually all accesses to that item will
return the last updated value.[1] Eventual consistency is
widely deployed in distributed systems, often under the
moniker of optimistic replication,[2] and has origins in
early mobile computing projects.[3] A system that has
achieved eventual consistency is often said to have converged, or achieved replica convergence.[4] Eventual
consistency is a weak guarantee - most stronger models,
like linearizability are trivially eventually consistent, but a
system that is merely eventually consistent doesn't usually
fulfill these stronger constraints.
Eventually consistent services are often classified
• Certified Administrative Professional, certification as providing BASE (Basically Available, Soft state,
from the International Association of Administra- Eventual consistency) semantics, in contrast to trative Professionals
ditional ACID (Atomicity, Consistency, Isolation,
Durability) guarantees.[5][6] Eventual consistency is
sometimes criticized[7] as increasing the complexity of
2.1.8 Other
distributed software applications. This is partly because
eventual consistency is purely a liveness guarantee (reads
• Carlos Andrés Pérez (1922-2010), twice President
eventually return the same value) and does not make
of Venezuela
safety guarantees: an eventually consistent system can
• CAP Markets, social franchise and supermarket return any value before it converges.
chain in Germany
• Capital Airlines, the ICAO airline designator for this 2.2.1 Conflict resolution
In order to ensure replica convergence, a system must rec• Causal adequacy principle, a philosophical claim oncile differences between multiple copies of distributed
made by René Descartes
data. This consists of two parts:
• Central Arizona Project, the Colorado River diversion canal in Arizona
• Chip Authentication Program, using EMV smartcards to authenticate online banking transactions
• Coded Anti-Piracy, an anti-piracy system for motion picture prints exhibited theatrically
• exchanging versions or updates of data between
servers (often known as anti-entropy);[8] and
• choosing an appropriate final state when concurrent
updates have occurred, called reconciliation.
The most appropriate approach to reconciliation depends
• Consolidated Appeals Process, a funding mecha- on the application. A widespread approach is “last writer
nism used by humanitarian aid organisations
wins”.[1] Another is to invoke a user-specified conflict
handler.[4] Timestamps and vector clocks are often used
to detect concurrency between updates.
Reconciliation of concurrent writes must occur sometime
before the next read, and can be scheduled at different
• Read repair: The correction is done when a read
finds an inconsistency. This slows down the read operation.
• Write repair: The correction takes place during a
write operation, if an inconsistency has been found,
slowing down the write operation.
• Asynchronous repair: The correction is not part of
a read or write operation.
Strong eventual consistency
[7] Yaniv Pessach (2013), Distributed Storage (Distributed
Storage: Concepts, Algorithms, and Implementations
ed.), Amazon, Systems using Eventual Consistency result
in decreased system load and increased system availability
but result in increased cognitive complexity for users and
[8] Demers, A.; Greene, D.; Hauser, C.; Irish, W.; Larson,
J. (1987). “Epidemic algorithms for replicated database
maintenance”. Proceedings of the sixth annual ACM Symposium on Principles of distributed computing - PODC '87.
p. 1. doi:10.1145/41840.41841. ISBN 978-0-89791239-6.
[9] Olivier Mallassi (2010-06-09). “Let’s play with Cassandra… (Part 1/3)". OCTO
Talks!. Retrieved 2011-03-23. Of course, at a given time,
chances are high that each node has its own version of the
data. Conflict resolution is made during the read requests
(called read-repair) and the current version of Cassandra
does not provide a Vector Clock conflict resolution mechanisms [sic] (should be available in the version 0.7). Conflict resolution is so based on timestamp (the one set when
you insert the row or the column): the higher timestamp
win[s] and the node you are reading the data [from] is responsible for that. This is an important point because the
timestamp is specified by the client, at the moment the
column is inserted. Thus, all Cassandra clients’ [sic] need
to be synchronized...
Whereas EC is only a liveness guarantee (updates will be
observed eventually), Strong Eventual Consistency (SEC)
adds the safety guarantee that any two nodes that have
received the same (unordered) set of updates will be in
the same state. If, furthermore, the system is monotonic,
the application will never suffer rollbacks. Conflict-free
replicated data types are a common approach to ensuring
[10] Shapiro, Marc; Preguiça, Nuno; Baquero, Carlos; ZaSEC.[10]
See also
• CAP theorem
[1] Vogels, W. (2009). “Eventually consistent”. Communications of the ACM 52: 40. doi:10.1145/1435417.1435432.
[2] Vogels, W. (2008). “Eventually Consistent”. Queue 6 (6):
14. doi:10.1145/1466443.1466448.
[3] Terry, D. B.; Theimer, M. M.; Petersen, K.; Demers, A.
J.; Spreitzer, M. J.; Hauser, C. H. (1995). “Managing
update conflicts in Bayou, a weakly connected replicated
storage system”. Proceedings of the fifteenth ACM symposium on Operating systems principles - SOSP '95. p. 172.
doi:10.1145/224056.224070. ISBN 0897917154.
[4] Petersen, K.; Spreitzer, M. J.; Terry, D. B.; Theimer,
M. M.; Demers, A. J. (1997).
“Flexible update
propagation for weakly consistent replication”. ACM
SIGOPS Operating Systems Review 31 (5):
[5] Pritchett, D. (2008). “Base: An Acid Alternative”. Queue
6 (3): 48. doi:10.1145/1394127.1394128.
[6] Bailis, P.; Ghodsi, A. (2013). “Eventual Consistency Today: Limitations, Extensions, and Beyond”. Queue 11 (3):
20. doi:10.1145/2460276.2462076.
wirski, Marek (2011-10-10). “Conflict-free replicated
data types”. SSS'11 Proceedings of the 13th international
conference on Stabilization, safety, and the security of
distributed systems (Springer-Verlag Berlin, Heidelberg):
2.3 Object-relational
The object-relational impedance mismatch is a set of
conceptual and technical difficulties that are often encountered when a relational database management system (RDBMS) is being used by a program written in
an object-oriented programming language or style, particularly when objects or class definitions are mapped
in a straightforward way to database tables or relational
The term object-relational impedance mismatch is derived
from the electrical engineering term impedance matching.
2.3.1 Mismatches
Object-oriented concepts
Encapsulation Object-oriented programs are designed
with techniques that result in encapsulated objects whose
representation is hidden. In an object-oriented framework, the underlying properties of a given object are ex-
pected to be unexposed to any interface outside of the
one implemented alongside the object. However, objectrelational mapping necessarily exposes the underlying
content of an object to interaction with an interface that
the object implementation cannot specify. Hence, objectrelational mapping violates the encapsulation of the object.
Accessibility In relational thinking, “private” versus
“public” access is relative to need rather than being an
absolute characteristic of the data’s state, as in the objectoriented (OO) model. The relational and OO models often have conflicts over relativity versus absolutism of classifications and characteristics.
Interface, class, inheritance and polymorphism Under an object-oriented paradigm, objects have interfaces
that together provide the only access to the internals of
that object. The relational model, on the other hand, utilizes derived relation variables (views) to provide varying perspectives and constraints to ensure integrity. Similarly, essential OOP concepts for classes of objects,
inheritance and polymorphism, are not supported by relational database systems.
els. In OO languages, objects can be composed of other
objects—often to a high degree—or specialize from a
more general definition. This may make the mapping to
relational schemas less straightforward. This is because
relational data tends to be represented in a named set
of global, unnested relation variables. Relations themselves, being sets of tuples all conforming to the same
header do not have an ideal counterpart in OO languages.
Constraints in OO languages are generally not declared as
such, but are manifested as exception raising protection
logic surrounding code that operates on encapsulated internal data. The relational model, on the other hand, calls
for declarative constraints on scalar types, attributes, relation variables, and the database as a whole.
Manipulative differences
The semantic differences are especially apparent in the
manipulative aspects of the contrasted models, however.
The relational model has an intrinsic, relatively small
and well-defined set of primitive operators for usage in
the query and manipulation of data, whereas OO languages generally handle query and manipulation through
custom-built or lower-level, case- and physical-accesspath-specific imperative operations. Some OO languages
Mapping to relational concepts A proper mapping do have support for declarative query sublanguages, but
between relational concepts and object-oriented concepts because OO languages typically deal with lists and percan be made if relational database tables are linked to haps hash tables, the manipulative primitives are necessarily distinct from the set-based operations of the relaassociations found in object-oriented analysis.
tional model.
Data type differences
A major mismatch between existing relational and OO
languages is the type system differences. The relational model strictly prohibits by-reference attributes (or
pointers), whereas OO languages embrace and expect
by-reference behavior. Scalar types and their operator
semantics can be vastly different between the models,
causing problems in mapping.
Transactional differences
The concurrency and transaction aspects are significantly
different also. In particular, transactions, the smallest
unit of work performed by databases, are much larger
in relational databases than are any operations performed
by classes in OO languages. Transactions in relational
databases are dynamically bounded sets of arbitrary data
manipulations, whereas the granularity of transactions in
an OO language is typically on the level of individual assignments to primitive-typed fields. In general, OO languages have no analogue of isolation or durability, so
atomicity and consistency are only ensured when writing
to fields of those primitive types.
For example, most SQL systems support string types
with varying collations and constrained maximum lengths
(open-ended text types tend to hinder performance),
while most OO languages consider collation only as an argument to sort routines and strings are intrinsically sized
to available memory. A more subtle, but related example
is that SQL systems often ignore trailing white space in a
string for the purposes of comparison, whereas OO string
libraries do not. It is typically not possible to construct
new data types as a matter of constraining the possible
values of other primitive types in an OO language.
Solving impedance mismatch
Solving the impedance mismatch problem for objectoriented programs starts with recognition of the differences in the specific logic systems being employed, then
Another mismatch has to do with the differences in the either the minimization or compensation of the misstructural and integrity aspects of the contrasted mod- match.
Structural and integrity differences
There have been some attempts at building objectoriented database management systems (OODBMS) that
would avoid the impedance mismatch problem. They
have been less successful in practice than relational
databases however, partly due to the limitations of OO
principles as a basis for a data model.[1] There has
been research performed in extending the database-like
capabilities of OO languages through such notions as
transactional memory.
The mixing of levels of discourse within OO application code presents problems, but there are some common mechanisms used to compensate. The biggest challenge is to provide framework support, automation of
data manipulation and presentation patterns, within the
level of discourse in which the domain data is being modeled. To address this, reflection and/or code generation
are utilized. Reflection allows code (classes) to be addressed as data and thus provide automation of the transport, presentation, integrity, etc. of the data. Generation addresses the problem through addressing the entity structures as data inputs for code generation tools or
meta-programming languages, which produce the classes
and supporting infrastructure en masse. Both of these
schemes may still be subject to certain anomalies where
these levels of discourse merge. For instance, generated
entity classes will typically have properties which map to
the domain (e. g. Name, Address) as well as properties
which provide state management and other framework infrastructure (e. g. IsModified).
One common solution to the impedance mismatch problem is to layer the domain and framework logic. In this
scheme, the OO language is used to model certain relational aspects at runtime rather than attempt the more
static mapping. Frameworks which employ this method
will typically have an analogue for a tuple, usually as a
“row” in a “dataset” component or as a generic “entity
instance” class, as well as an analogue for a relation. Advantages of this approach may include:
• Straightforward paths to build frameworks and automation around transport, presentation, and validation of domain data.
2.3.3 Contention
• Smaller code size; faster compile and load times.
It has been argued, by Christopher J. Date and others, that a truly relational DBMS would pose no such
problem,[3][4][5] as domains and classes are essentially one
• Avoids the name-space and semantic mismatch is- and the same thing. A naïve mapping between classes and
relational schemata is a fundamental design mistake ; and
that individual tuples within a database table (relation)
• Expressive constraint checking
ought to be viewed as establishing relationships between
entities; not as representations for complex entities them• No complex mapping necessary
selves. However, this view tends to diminish the influence
and role of object-oriented programming, using it as little
Disadvantages may include:
more than a field type management system.
• Ability for the schema to change dynamically.
The impedance mismatch is in programming between the
• Lack of static type “safety” checks. Typed accessors
domain objects and the user interface. Sophisticated user
are sometimes utilized as one way to mitigate this.
interfaces, to allow operators, managers, and other non• Possible performance cost of runtime construction programmers to access and manipulate the records in the
database, often require intimate knowledge about the naand access.
ture of the various database attributes (beyond name and
• Inability to natively utilize uniquely OO aspects, type). In particular, it’s considered a good practice (from
an end-user productivity point of view) to design user
such as polymorphism.
interfaces such that the UI prevents illegal transactions
(those which cause a database constraint to be violated)
Alternative architectures
from being entered; to do so requires much of the logic
present in the relational schemata to be duplicated in the
The rise of XML databases and XML client structures has code.
motivated other alternative architectures to get around
the impedance mismatch challenges. These architectures Certain code-development frameworks can leverage ceruse XML technology in the client (such as XForms) and tain forms of logic that are represented in the database’s
native XML databases on the server that use the XQuery schema (such as referential integrity constraints), so that
language for data selection. This allows a single data such issues are handled in a generic and standard fashion
model and a single data selection language (XPath) to be through library routines rather than ad hoc code written
used in the client, in the rules engines and on the persis- on a case-by-case basis.
tence server.[2]
It has been argued that SQL, due to a very limited set of
domain types (and other alleged flaws) makes proper object and domain-modelling difficult; and that SQL constitutes a very lossy and inefficient interface between a
DBMS and an application program (whether written in an
object-oriented style or not). However, SQL is currently
the only widely accepted common database language in
the marketplace; use of vendor-specific query languages
is seen as a bad practice when avoidable. Other database
languages such as Business System 12 and Tutorial D
have been proposed; but none of these has been widely
adopted by DBMS vendors.
In current versions of mainstream “object-relational”
DBMSs like Oracle and Microsoft SQL Server, the above
point may be a non-issue. With these engines, the functionality of a given database can be arbitrarily extended
through stored code (functions and procedures) written
in a modern OO language (Java for Oracle, and a Microsoft .NET language for SQL Server), and these functions can be invoked in-turn in SQL statements in a transparent fashion: that is, the user neither knows nor cares
that these functions/procedures were not originally part
of the database engine. Modern software-development
paradigms are fully supported: thus, one can create a set
of library routines that can be re-used across multiple
database schemas.
These vendors decided to support OO-language integration at the DBMS back-end because they realized
that, despite the attempts of the ISO SQL-99 committee to add procedural constructs to SQL, SQL will never
have the rich set of libraries and data structures that
today’s application programmers take for granted, and
it is reasonable to leverage these as directly as possible rather than attempting to extend the core SQL language. Consequently, the difference between “application programming” and “database administration” is now
blurred: robust implementation of features such as constraints and triggers may often require an individual with
dual DBA/OO-programming skills, or a partnership between individuals who combine these skills. This fact also
bears on the “division of responsibility” issue below.
held by an application program are just that — temporary copies (which may be out of date, if the underlying
database record was subsequently modified by a transaction). Many object-oriented programmers prefer to view
the in-memory representations of objects themselves as
the canonical data, and view the database as a backing
store and persistence mechanism.
Another point of contention is the proper division
of responsibility between application programmers and
database administrators (DBA). It is often the case that
needed changes to application code (in order to implement a requested new feature or functionality) require
corresponding changes in the database definition; in most
organizations, the database definition is the responsibility of the DBA. Due to the need to maintain a production database system 24 hours a day many DBAs are reluctant to make changes to database schemata that they
deem gratuitous or superfluous and in some cases outright
refuse to do so. Use of developmental databases (apart
from production systems) can help somewhat; but when
the newly developed application “goes live” the DBA
will need to approve any changes. Some programmers
view this as intransigence; however the DBA is frequently
held responsible if any changes to the database definition
cause a loss of service in a production system—as a result,
many DBAs prefer to contain design changes to application code, where design defects are far less likely to have
catastrophic consequences.
In organizations with a non-dysfunctional relationship between DBAs and developers, though, the above issue
should not present itself, as the decision to change a
database schema or not would only be driven by business
needs: a new requirement to persist additional data or a
performance boost of a critical application would both
trigger a schema modification, for example.
2.3.4 Philosophical differences
Key philosophical differences between the OO and relaSome, however, would point out that this contention is tional models can be summarized as follows:
moot due to the fact that: (1) RDBMSes were never intended to facilitate object modelling, and (2) SQL gen• Declarative vs. imperative interfaces — Relaerally should only be seen as a “lossy” or “inefficient”
tional thinking tends to use data as interfaces, not
interface language when one is trying to achieve a sobehavior as interfaces. It thus has a declarative tilt
lution for which RDBMSes were not designed. SQL is
in design philosophy in contrast to OO’s behavioral
very efficient at doing what it was designed to do, namely,
tilt. (Some relational proponents propose using trigto query, sort, filter, and store large sets of data. Some
gers, stored procedures, etc. to provide complex bewould additionally point out that the inclusion of OO lanhavior, but this is not a common viewpoint.)
guage functionality in the back-end simply facilitates bad
architectural practice, as it admits high-level application
• Schema bound — Objects do not have to follow
logic into the data tier, antithetical to the RDBMS.
a “parent schema” for which attributes or accessors
an object has, while table rows must follow the enHere the “canonical” copy of state is located. The
tity’s schema. A given row must belong to one and
database model generally assumes that the database manonly one entity. The closest thing in OO is inheragement system is the only authoritative repository of
itance, but it is generally tree-shaped and optional.
state concerning the enterprise; any copies of such state
A dynamic form of relational tools that allows ad
hoc columns may relax schema bound-ness, but such
tools are currently rare.
• Access rules — In relational databases, attributes
are accessed and altered through predefined relational operators, while OO allows each class to create its own state alteration interface and practices.
The “self-handling noun” viewpoint of OO gives independence to each object that the relational model
does not permit. This is a “standards versus local
freedom” debate. OO tends to argue that relational
standards limit expressiveness, while relational proponents suggest the rule adherence allows more abstract math-like reasoning, integrity, and design consistency.
• Relationship between nouns and verbs — OO encourages a tight association between verbs (actions)
and the nouns (entities) that the operations operate
on. The resulting tightly bound entity containing
both nouns and the verbs is usually called a class, or
in OO analysis, a concept. Relational designs generally do not assume there is anything natural or logical
about such tight associations (outside of relational
• Object identity — Objects (other than immutable
ones) are generally considered to have a unique identity; two objects which happen to have the same
state at a given point in time are not considered to
be identical. Relations, on the other hand, have no
inherent concept of this kind of identity. That said,
it is a common practice to fabricate “identity” for
records in a database through use of globally unique
candidate keys; though many consider this a poor
practice for any database record which does not have
a one-to-one correspondence with a real world entity. (Relational, like objects, can use domain keys
if they exist in the external world for identification
purposes). Relational systems in practice strive for
and support “permanent” and inspectable identification techniques, whereas object identification techniques tend to be transient or situational.
• Normalization — Relational normalization practices are often ignored by OO designs. However,
this may just be a bad habit instead of a native feature of OO. An alternate view is that a collection
of objects, interlinked via pointers of some sort, is
equivalent to a network database; which in turn can
be viewed as an extremely denormalized relational
• Schema inheritance — Most relational databases
do not support schema inheritance. Although such a
feature could be added in theory to reduce the conflict with OOP, relational proponents are less likely
to believe in the utility of hierarchical taxonomies
and sub-typing because they tend to view set-based
taxonomies or classification systems as more powerful and flexible than trees. OO advocates point out
that inheritance/subtyping models need not be limited to trees (though this is a limitation in many popular OO languages such as Java), but non-tree OO
solutions are seen as more difficult to formulate than
set-based variation-on-a-theme management techniques preferred by relational. At the least, they differ from techniques commonly used in relational algebra.
• Structure vs. behaviour — OO primarily focuses
on ensuring that the structure of the program is
reasonable (maintainable, understandable, extensible, reusable, safe), whereas relational systems focus on what kind of behaviour the resulting run-time
system has (efficiency, adaptability, fault-tolerance,
liveness, logical integrity, etc.). Object-oriented
methods generally assume that the primary user of
the object-oriented code and its interfaces are the application developers. In relational systems, the endusers’ view of the behaviour of the system is sometimes considered to be more important. However,
relational queries and “views” are common techniques to present information in application- or taskspecific configurations. Further, relational does not
prohibit local or application-specific structures or tables from being created, although many common development tools do not directly provide such a feature, assuming objects will be used instead. This
makes it difficult to know whether the stated nondeveloper perspective of relational is inherent to relational, or merely a product of current practice and
tool implementation assumptions.
• Set vs. graph relationships — The relationship
between different items (objects or records) tend to
be handled differently between the paradigms. Relational relationships are usually based on idioms
taken from set theory, while object relationships
lean toward idioms adopted from graph theory (including trees). While each can represent the same
information as the other, the approaches they provide to access and manage information differ.
As a result of the object-relational impedance mismatch,
it is often argued by partisans on both sides of the debate that the other technology ought to be abandoned or
reduced in scope.[6] Some database advocates view traditional “procedural” languages as more compatible with
an RDBMS than many OO languages; or suggest that a
less OO-style ought to be used. (In particular, it is argued that long-lived domain objects in application code
ought not to exist; any such objects that do exist should
be created when a query is made and disposed of when
a transaction or task is complete). On the other hand,
many OO advocates argue that more OO-friendly persistence mechanisms, such as OODBMS, ought to be developed and used, and that relational technology ought
to be phased out. Of course, it should be pointed out
that many (if not most) programmers and DBAs do not
hold either of these viewpoints; and view the objectrelational impedance mismatch as a mere fact of life that
information technology has to deal with.
It is also argued that the O/R mapping is paying off in
some situations, but is probably oversold: it has advantages besides drawbacks. Skeptics point out that it is
worth to think carefully before using it, as it will add little
value in some cases.[7]
Object-Oriented Model
Object 1: Maintenance Report
Activity Code
Route No.
Daily Production
Equipment Hours
Labor Hours
Object 1 Instance
Object 2: Maintenance Activity
Activity Code
Activity Name
Production Unit
Average Daily Production Rate
[1] C. J. Date, Relational Database Writings
[2] Dan McCreary, XRX: Simple, Elegant, Disruptive on
[3] Date, Christopher ‘Chris’ J; Pascal, Fabian (2012-08-12)
[2005], “Type vs. Domain and Class”, Database debunkings (World Wide Web log), Google, retrieved 12 September 2012.
Example of an object-oriented model[1]
Object databases have been considered since the early
2.4.1 Overview
[4] ——— (2006), “4. On the notion of logical difference”, Date on Database: writings 2000–2006, The expert’s voice in database; Relational database select writings, USA: Apress, p. 39, ISBN 978-1-59059-746-0,
Class seems to be indistinguishable from type, as that term
is classically understood.
(OODBMSs) combine database capabilities with
object-oriented programming language capabilities.
OODBMSs allow object-oriented programmers to develop the product, store them as objects, and replicate or
modify existing objects to make new objects within the
[5] ——— (2004), “26. Object/Relational databases”, An in- OODBMS. Because the database is integrated with the
troduction to database systems (8th ed.), Pearson Addison
programming language, the programmer can maintain
Wesley, p. 859, ISBN 978-0-321-19784-9, ...any such
consistency within one environment, in that both the
rapprochement should be firmly based on the relational
OODBMS and the programming language will use
the same model of representation. Relational DBMS
[6] Neward, Ted (2006-06-26). “The Vietnam of Computer projects, by way of contrast, maintain a clearer division
Science”. Interoperability Happens. Retrieved 2010-06- between the database model and the application.
As the usage of web-based technology increases with
the implementation of Intranets and extranets, companies
have a vested interest in OODBMSs to display their complex data. Using a DBMS that has been specifically designed to store data as objects gives an advantage to those
2.3.6 External links
companies that are geared towards multimedia presentathat utilize computer-aided design
• The Object-Relational Impedance Mismatch - Agile tion or organizations
Data Essay
[7] J2EE Design and Development by Rod Johnson, © 2002
Wrox Press, p. 256.
Some object-oriented databases are designed to work
• The Vietnam of Computer Science - Examples of
well with object-oriented programming languages such
mismatch problems
as Delphi, Ruby, Python, Perl, Java, C#, Visual Basic
.NET, C++, Objective-C and Smalltalk; others have their
own programming languages. OODBMSs use exactly the
2.4 Object database
same model as object-oriented programming languages.
An object database (also object-oriented database
management system) is a database management system in which information is represented in the form of
objects as used in object-oriented programming. Object
databases are different from relational databases which
are table-oriented. Object-relational databases are a hybrid of both approaches.
2.4.2 History
Object database management systems grew out of research during the early to mid-1970s into having intrinsic database management support for graph-structured
objects. The term “object-oriented database system”
first appeared around 1985.[4] Notable research projects
included Encore-Ob/Server (Brown University), EXODUS (University of Wisconsin–Madison), IRIS (HewlettPackard), ODE (Bell Labs), ORION (Microelectronics
and Computer Technology Corporation or MCC), Vodak (GMD-IPSI), and Zeitgeist (Texas Instruments). The
ORION project had more published papers than any of
the other efforts. Won Kim of MCC compiled the best
of those papers in a book published by The MIT Press.[5]
Early commercial products included Gemstone (Servio
Logic, name changed to GemStone Systems), Gbase
(Graphael), and Vbase (Ontologic). The early to mid1990s saw additional commercial products enter the market. These included ITASCA (Itasca Systems), Jasmine
(Fujitsu, marketed by Computer Associates), Matisse
(Matisse Software), Objectivity/DB (Objectivity, Inc.),
ObjectStore (Progress Software, acquired from eXcelon
which was originally Object Design), ONTOS (Ontos,
Inc., name changed from Ontologic), O2 [6] (O2 Technology, merged with several companies, acquired by
Informix, which was in turn acquired by IBM), POET
(now FastObjects from Versant which acquired Poet Software), Versant Object Database (Versant Corporation),
VOSS (Logic Arts) and JADE (Jade Software Corporation). Some of these products remain on the market and
have been joined by new open source and commercial
products such as InterSystems Caché.
Object database management systems added the concept
of persistence to object programming languages. The
early commercial products were integrated with various
languages: GemStone (Smalltalk), Gbase (LISP), Vbase
(COP) and VOSS (Virtual Object Storage System for
Smalltalk). For much of the 1990s, C++ dominated the
commercial object database management market. Vendors added Java in the late 1990s and more recently, C#.
Starting in 2004, object databases have seen a second growth period when open source object databases
emerged that were widely affordable and easy to use,
because they are entirely written in OOP languages
like Smalltalk, Java, or C#, such as Versant’s db4o
(db4objects), DTS/S1 from Obsidian Dynamics and Perst
(McObject), available under dual open source and commercial licensing.
• 1982
• Gemstone started (as Servio Logic) to build a
set theoretic model data base machine.
• 1985 – Term Object Database first introduced
• 1986
• Servio Logic (Gemstone Systems) Ships Gemstone 1.0
• 1988
• Versant Corporation started (as Object Sciences Corp)
• Objectivity, Inc. founded
• Early 1990s
• Servio Logic changes name to Gemstone Systems
• Gemstone (Smalltalk)-(C++)-(Java)
• GBase (LISP)
• Objectivity/DB
• Mid 1990’s
• InterSystems Caché
• Versant Object Database
• ObjectStore
• Poet
• Jade
• Matisse
• Illustra Informix
• Webcrossing
• 2000’s
• db4o project started by Carl Rosenberger
• ObjectDB
• 2001 IBM acquires Informix
• 1966
• 1979
• InterSystems M
• 1980
• TORNADO – an object database for
• 2003 odbpp public release
• 2004 db4o’s commercial launch as db4objects, Inc.
• 2008 db4o acquired by Versant Corporation
• 2010 VMware acquires GemStone[8]
• 2012 Wakanda first production versions with open
source and commercial licenses
• 2013 GemTalk Systems acquires GemStone products from VMware[9]
• 2014 Realm
Adoption of object databases
Object databases based on persistent programming acquired a niche in application areas such as engineering
and spatial databases, telecommunications, and scientific
areas such as high energy physics and molecular biology.
Another group of object databases focuses on embedded
use in devices, packaged software, and real-time systems.
Technical features
Most object databases also offer some kind of query language, allowing objects to be found using a declarative
programming approach. It is in the area of object query
languages, and the integration of the query and navigational interfaces, that the biggest differences between
products are found. An attempt at standardization was
made by the ODMG with the Object Query Language,
ODMG 3.0. By 2001, most of the major object database
and object-relational mapping vendors claimed conformance to the ODMG Java Language Binding. Compliance to the other components of the specification was
mixed. In 2001, the ODMG Java Language Binding was
submitted to the Java Community Process as a basis for
the Java Data Objects specification. The ODMG member companies then decided to concentrate their efforts
on the Java Data Objects specification. As a result, the
ODMG disbanded in 2001.
Many object database ideas were also absorbed into SQL:
1999 and have been implemented in varying degrees in
object-relational database products.
In 2005 Cook, Rai, and Rosenberger proposed to drop
all standardization efforts to introduce additional objectoriented query APIs but rather use the OO programming
language itself, i.e., Java and .NET, to express queries.
As a result, Native Queries emerged. Similarly, Microsoft announced Language Integrated Query (LINQ)
and DLINQ, an implementation of LINQ, in SeptemAccess to data can be faster because joins are often not ber 2005, to provide close, language-integrated database
needed (as in a tabular implementation of a relational query capabilities with its programming languages C#
database). This is because an object can be retrieved di- and VB.NET 9.
rectly without a search, by following pointers.
In February 2006, the Object Management Group
Another area of variation between products is in the way
that the schema of a database is defined. A general characteristic, however, is that the programming language and
the database schema use the same type definitions.
Multimedia applications are facilitated because the class
methods associated with the data are responsible for its
correct interpretation.
Many object databases, for example Gemstone or VOSS,
offer support for versioning. An object can be viewed
as the set of all its versions. Also, object versions can
be treated as objects in their own right. Some object
databases also provide systematic support for triggers and
constraints which are the basis of active databases.
The efficiency of such a database is also greatly improved
in areas which demand massive amounts of data about
one item. For example, a banking institution could get the
user’s account information and provide them efficiently
with extensive information such as transactions, account
information entries etc. The Big O Notation for such a
database paradigm drops from O(n) to O(1), greatly increasing efficiency in these specific cases.
The Object Data Management Group was a consortium
of object database and object-relational mapping vendors, members of the academic community, and interested parties. Its goal was to create a set of specifications that would allow for portable applications that store
objects in database management systems. It published
several versions of its specification. The last release was
(OMG) announced that they had been granted the right
to develop new specifications based on the ODMG 3.0
specification and the formation of the Object Database
Technology Working Group (ODBT WG). The ODBT
WG planned to create a set of standards that would incorporate advances in object database technology (e.g.,
replication), data management (e.g., spatial indexing),
and data formats (e.g., XML) and to include new features into these standards that support domains where object databases are being adopted (e.g., real-time systems).
The work of the ODBT WG was suspended in March
2009 when, subsequent to the economic turmoil in late
2008, the ODB vendors involved in this effort decided to
focus their resources elsewhere.
In January 2007 the World Wide Web Consortium gave
final recommendation status to the XQuery language.
XQuery uses XML as its data model. Some of the ideas
developed originally for object databases found their
way into XQuery, but XQuery is not intrinsically objectoriented. Because of the popularity of XML, XQuery
engines compete with object databases as a vehicle for
storage of data that is too complex or variable to hold
conveniently in a relational database. XQuery also allows
modules to be written to provide encapsulation features
that have been provided by Object-Oriented systems.
2.4.7 Comparison with RDBMSs
An object database stores complex data and relationships
between data directly, without mapping to relational rows
and columns, and this makes them suitable for applications dealing with very complex data.[10] Objects have a
many to many relationship and are accessed by the use of
pointers. Pointers are linked to objects to establish relationships. Another benefit of an OODBMS is that it can
be programmed with small procedural differences without affecting the entire system.[11]
See also
• Comparison of object database management systems
• Component-oriented database
[8] “SpringSource to Acquire Gemstone Systems Data Management Technology”. WMware. May 6, 2010. Retrieved August 5, 2014.
[9] GemTalk Systems (May 2, 2013). “GemTalk Systems
Acquires GemStone/S Products from VMware”. PRWeb.
Retrieved August 5, 2014.
[10] Radding, Alan (1995). “So what the Hell is ODBMS?".
Computerworld 29 (45): 121–122, 129.
[11] Burleson, Donald. (1994). OODBMSs gaining MIS
ground but RDBMSs still own the road. Software Magazine, 14(11), 63
• EDA database
• Enterprise Objects Framework
2.4.10 External links
• Object DBMS resource portal
• Object Data Management Group
• Object-Oriented Databases – From
• Object-relational database
• Persistence (computer science)
• DB-Engines Ranking of Object Oriented DBMS by
popularity, updated monthly
• Relational model
[1] Data Integration Glossary, U.S. Department of Transportation, August 2001.
[2] ODBMS.ORG :: Object Database (ODBMS) | ObjectOriented Database (OODBMS) | Free Resource Portal. ODBMS (2013-08-31). Retrieved on 2013-09-18.
Archived July 25, 2014 at the Wayback Machine
[3] O’Brien, J. A., & Marakas, G. M. (2009). Management
Information Systems (9th ed.). New York, NY: McGrawHill/Irwin
[4] Three example references from 1985 that use the term:
T. Atwood, “An Object-Oriented DBMS for Design Support Applications,” Proceedings of the IEEE COMPINT 85,
pp. 299-307, September 1985; N. Derrett, W. Kent, and
P. Lyngbaek, “Some Aspects of Operations in an ObjectOriented Database,” Database Engineering, vol. 8, no.
4, IEEE Computer Society, December 1985; D. Maier,
A. Otis, and A. Purdy, “Object-Oriented Database Development at Servio Logic,” Database Engineering, vol. 18,
no.4, December 1985.
[5] Kim, Won. Introduction to Object-Oriented Databases.
The MIT Press, 1990. ISBN 0-262-11124-1
[6] Bancilhon, Francois; Delobel,Claude; and Kanellakis,
Paris. Building an Object-Oriented Database System: The
Story of O2 . Morgan Kaufmann Publishers, 1992. ISBN
[7] Ulfsby; et al. (July 1981). “TORNADO: a DBMS for
CAD/CAM systems”. Computer-Aided Design 13 (4):
2.5 NoSQL
“Structured storage” redirects here. For the Microsoft
technology also known as structured storage, see COM
Structured Storage.
A NoSQL (originally referring to “non SQL” or “non relational” [1] ) database provides a mechanism for storage
and retrieval of data that is modeled in means other than
the tabular relations used in relational databases. Such
databases have existed since the late 1960s, but did not
obtain the “NoSQL” moniker until a surge of popularity in the early twenty-first century,[2] triggered by the
needs of Web 2.0 companies such as Facebook, Google
Motivations for this approach include: simplicity of design, simpler “horizontal” scaling to clusters of machines,
which is a problem for relational databases,[2] and finer
control over availability. The data structures used by
NoSQL databases (e.g. key-value, wide column, graph,
or document) differ slightly from those used by default
in relational databases, making some operations faster in
NoSQL and others faster in relational databases. The particular suitability of a given NoSQL database depends on
the problem it must solve. Sometimes the data structures
used by NoSQL databases are also viewed as “more flexible” than relational database tables.[6]
NoSQL databases are increasingly used in big data and
real-time web applications.[7] NoSQL systems are also
sometimes called “Not only SQL” to emphasize that they
may support SQL-like query languages.[8][9]
2.5. NOSQL
Many NoSQL stores compromise consistency (in the
sense of the CAP theorem) in favor of availability, partition tolerance, and speed. Barriers to the greater adoption of NoSQL stores include the use of low-level query
languages (instead of SQL, for instance the lack of ability to perform ad-hoc JOIN’s across tables), lack of standardized interfaces, and huge previous investments in existing relational databases.[10] Most NoSQL stores lack
true ACID transactions, although a few databases, such
as MarkLogic, Aerospike, FairCom c-treeACE, Google
Spanner (though technically a NewSQL database), Symas
LMDB and OrientDB have made them central to their designs. (See ACID and JOIN Support.)
are MongoDB, Apache Cassandra, and Redis.[20]
2.5.2 Types and examples of NoSQL
There have been various approaches to classify NoSQL
databases, each with different categories and subcategories, some of which overlap. A basic classification
based on data model, with examples:
• Column: Accumulo, Cassandra, Druid, HBase,
Instead, most NoSQL databases offer a concept of “even• Document:
Apache CouchDB, Clusterpoint,
tual consistency” in which database changes are propaCouchbase, DocumentDB, HyperDex, Lotus
gated to all nodes “eventually” (typically within millisecNotes, MarkLogic, MongoDB, OrientDB, Qizx
onds) so queries for data might not return updated data
• Key-value:
Aerospike, CouchDB, Dynamo,
immediately or might result in reading data that is not acFairCom c-treeACE, FoundationDB, HyperDex,
curate, a problem known as stale reads.[11] Additionally,
MemcacheDB, MUMPS, Oracle NoSQL Database,
some NoSQL systems may exhibit lost writes and other
OrientDB, Redis, Riak
forms of data loss.[12] Fortunately, some NoSQL systems
provide concepts such as write-ahead logging to avoid
• Graph: Allegro, InfiniteGraph, MarkLogic, Neo4J,
data loss.[13] For distributed transaction processing across
OrientDB, Virtuoso, Stardog
multiple databases, data consistency is an even bigger
challenge that is difficult for both NoSQL and relational
• Multi-model: Alchemy Database, ArangoDB, Cordatabases. Even current relational databases “do not altexDB, FoundationDB, MarkLogic, OrientDB
low referential integrity constraints to span databases.”[14]
There are few systems that maintain both ACID transacA more detailed classification is the following, based on
tions and X/Open XA standards for distributed transacone from Stephen Yen:[21]
tion processing.
Correlation databases are model-independent, and instead of row-based or column-based storage, use valuebased storage.
2.5.1 History
The term NoSQL was used by Carlo Strozzi in 1998 to
name his lightweight, Strozzi NoSQL open-source relational database that did not expose the standard SQL interface, but was still relational.[15] His NoSQL RDBMS is
distinct from the circa-2009 general concept of NoSQL
databases. Strozzi suggests that, as the current NoSQL
movement “departs from the relational model altogether;
it should therefore have been called more appropriately
'NoREL'",[16] referring to 'No Relational'.
Johan Oskarsson of reintroduced the term
NoSQL in early 2009 when he organized an event
to discuss “open source distributed, non relational
databases".[17] The name attempted to label the emergence of an increasing number of non-relational, distributed data stores, including open source clones of
Google’s BigTable/MapReduce and Amazon’s Dynamo.
Most of the early NoSQL systems did not attempt to provide atomicity, consistency, isolation and durability guarantees, contrary to the prevailing practice among relational database systems.[18]
Key-value stores
Main article: Key-value database
Key-value (KV) stores use the associative array (also
known as a map or dictionary) as their fundamental data
model. In this model, data is represented as a collection
of key-value pairs, such that each possible key appears at
most once in the collection.[22][23]
The key-value model is one of the simplest non-trivial
data models, and richer data models are often implemented on top of it. The key-value model can be
extended to an ordered model that maintains keys in
lexicographic order. This extension is powerful, in that
it can efficiently process key ranges.[24]
Key-value stores can use consistency models ranging
from eventual consistency to serializability. Some support ordering of keys. Some maintain data in memory
(RAM), while others employ solid-state drives or rotating
Based on 2014 revenue, the NoSQL market leaders are
MarkLogic, MongoDB, and Datastax.[19] Based on 2015 Examples include Oracle NoSQL Database, redis, and
popularity rankings, the most popular NoSQL databases dbm.
Document store
Main articles: Document-oriented database and XML
The central concept of a document store is the notion of
a “document”. While each document-oriented database
implementation differs on the details of this definition, in
general, they all assume that documents encapsulate and
encode data (or information) in some standard formats or
encodings. Encodings in use include XML, YAML, and
JSON as well as binary forms like BSON. Documents are
addressed in the database via a unique key that represents
that document. One of the other defining characteristics
of a document-oriented database is that in addition to the
key lookup performed by a key-value store, the database
offers an API or query language that retrieves documents
based on their contents
• InterSystems Caché
• NeoDatis ODB
• ObjectDatabase++
• ObjectDB
• Objectivity/DB
• ObjectStore
• Perst
• OpenLink Virtuoso
• Versant Object Database
Different implementations offer different ways of orgaTabular
nizing and/or grouping documents:
• Collections
• Tags
• Non-visible metadata
• Directory hierarchies
• Apache Accumulo
• BigTable
• Apache Hbase
• Hypertable
• Mnesia
• OpenLink Virtuoso
Compared to relational databases, for example, collections could be considered analogous to tables and documents analogous to records. But they are different: every Tuple store
record in a table has the same sequence of fields, while
documents in a collection may have fields that are com• Apache River
pletely different.
• GigaSpaces
Main article: Graph database
• Tarantool
• TIBCO ActiveSpaces
• OpenLink Virtuoso
This kind of database is designed for data whose relations are well represented as a graph (elements intercon- Triple/quad store (RDF) database
nected with an undetermined number of relations between them). The kind of data could be social relations, Main articles: Triplestore and Named graph
public transport links, road maps or network topologies,
for example.
• AllegroGraph
Graph databases and their query language
• Apache JENA (It’s a framework, not a database)
• MarkLogic
Object database
• Ontotext-OWLIM
Main article: Object database
• Oracle NoSQL database
• SparkleDB
• db4o
• Virtuoso Universal Server
• GemStone/S
• Stardog
2.5. NOSQL
• Amazon DynamoDB
• Amazon SimpleDB
• Datastore on Google Appengine
• Clusterpoint database
• Cloudant Data Layer (CouchDB)
• Freebase
• Microsoft Azure Tables [25]
• Microsoft Azure DocumentDB [26]
• OpenLink Virtuoso
2.5.4 Handling relational data
Since most NoSQL databases lack ability for joins in
queries, the database schema generally needs to be designed differently. There are three main techniques for
handling relational data in a NoSQL database. (See table
Join and ACID Support for NoSQL databases that support joins.)
Multiple queries
Instead of retrieving all the data with one query, it’s common to do several queries to get the desired data. NoSQL
queries are often faster than traditional SQL queries so
the cost of having to do additional queries may be acceptable. If an excessive number of queries would be
necessary, one of the other two approaches is more appropriate.
Multivalue databases
• D3 Pick database
Caching/replication/non-normalized data
Instead of only storing foreign keys, it’s common to store
actual foreign values along with the model’s data. For example, each blog comment might include the username
in addition to a user id, thus providing easy access to
the username without requiring another lookup. When
InterSystems Caché
a username changes however, this will now need to be
Northgate Information Solutions Reality, the origi- changed in many places in the database. Thus this apnal Pick/MV Database
proach works better when reads are much more common
than writes.[28]
• Extensible Storage Engine (ESE/NT)
• Revelation Software’s OpenInsight
• Rocket U2
Multimodel database
• OrientDB
• FoundationDB
Nesting data
With document databases like MongoDB it’s common to
put more data in a smaller number of collections. For
example, in a blogging application, one might choose to
store comments within the blog post document so that
with a single retrieval one gets all the comments. Thus
in this approach a single document contains all the data
you need for a specific task.
• ArangoDB
• MarkLogic
2.5.5 ACID and JOIN Support
If a database is marked as supporting ACID or joins, then
the documentation for the database makes that claim.
2.5.3 Performance
The degree to which the capability is fully supported in a
manner similar to most SQL databases or the degree to
Ben Scofield rated different categories of NoSQL which it meets the needs of a specific application is left
databases as follows: [27]
up to the reader to assess.
Performance and scalability comparisons are sometimes (*) HyperDex currently offers ACID support via its Warp
done with the YCSB benchmark.
extension, which is a commercial add-on. (**) Joins do
See also: Comparison of structured storage software
not necessarily apply to document databases, but MarkLogic can do joins using semantics [29]
See also
[12] Martin Zapletal: Large volume data analysis on the Typesafe Reactive Platform, ScalaDays 2015, Slides
• CAP theorem
• Comparison of object database management systems
• Correlation database
[15] Lith, Adam; Mattson, Jakob (2010). “Investigating storage solutions for large data: A comparison of well performing and scalable data storage solutions for real time
extraction and batch insertion of data” (PDF). Göteborg: Department of Computer Science and Engineering,
Chalmers University of Technology. p. 70. Retrieved
12 May 2011. Carlo Strozzi first used the term NoSQL
in 1998 as a name for his open source relational database
that did not offer a SQL interface[...]
• Distributed cache
• Faceted search
• MultiValue database
• Multi-model database
• Triplestore
[16] “NoSQL Relational Database Management System:
Home Page”. 2 October 2007. Retrieved 29
March 2010.
Generation Databases mostly addressing some of the
points: being non-relational, distributed, open-source and
horizontally scalable”
[2] Leavitt, Neal (2010). “Will NoSQL Databases Live Up
to Their Promise?" (PDF). IEEE Computer.
[3] Mohan, C. (2013). History Repeats Itself: Sensible and
NonsenSQL Aspects of the NoSQL Hoopla (PDF). Proc.
16th Int'l Conf. on Extending Database Technology.
and BigTables”
• Comparison of structured storage software
lose data” section
“Amazon helped start the “NoSQL” movement.”
amazon-dynamodb.html “Customers like SimpleDB’s
table interface and its flexible data model. Not having to
update their schemas when their systems evolve makes
life much easier”
[7] “RDBMS dominate the database market, but NoSQL systems are catching up”. 21 Nov 2013.
Retrieved 24 Nov 2013.
[8] “NoSQL (Not Only SQL)". NoSQL database, also called
Not Only SQL
[9] Fowler, Martin. “NosqlDefinition”. many advocates of
NoSQL say that it does not mean a “no” to SQL, rather it
means Not Only SQL
[10] Grolinger, K.; Higashino, W. A.; Tiwari, A.; Capretz,
M. A. M. (2013). “Data management in cloud environments: NoSQL and NewSQL data stores” (PDF). JoCCASA, Springer. Retrieved 8 Jan 2014.
[17] “NoSQL 2009”. 12 May 2009. Retrieved 29 March 2010.
[18] Chapple, Mike. “The ACID Model”.
[19] “Hadoop-NoSQL-rankings”. Retrieved 2015-11-17.
[20] “DB-Engines Ranking”. Retrieved 2015-07-31.
[21] Yen, Stephen. “NoSQL is a Horseless Carriage” (PDF).
NorthScale. Retrieved 2014-06-26..
[22] Sandy (14 January 2011). “Key Value stores and the
NoSQL movement”.
Stackexchange. Retrieved 1 January 2012. Key-value
stores allow the application developer to store schema-less
data. This data usually consists of a string that represents
the key, and the actual data that is considered the value
in the “key-value” relationship. The data itself is usually
some kind of primitive of the programming language (a
string, an integer, or an array) or an object that is being
marshaled by the programming language’s bindings to
the key-value store. This structure replaces the need for
a fixed data model and allows proper formatting.
[23] Seeger, Marc (21 September 2009). “Key-Value Stores:
a practical overview” (PDF). http://blog.marc-seeger.
Marc Seeger. Retrieved 1 January 2012. Key-value
stores provide a high-performance alternative to relational
database systems with respect to storing and accessing
data. This paper provides a short overview of some of the
currently available key-value stores and their interface to
the Ruby programming language.
[24] Katsov, Ilya (1 March 2012). “NoSQL Data Modeling
Techniques”. Ilya Katsov. Retrieved 8 May 2014.
[27] Scofield, Ben (2010-01-14). “NoSQL - Death to Relational Databases(?)". Retrieved 2014-06-26.
[28] “Making the Shift from Relational to NoSQL” (PDF). Retrieved December 5, 2014.
Further reading
A tabular data card proposed for Babbage’s Analytical Engine
• Sadalage, Pramod; Fowler, Martin (2012). NoSQL showing a key-value pair, in this instance a number and its baseDistilled: A Brief Guide to the Emerging World of ten logarithm.
Polyglot Persistence. Addison-Wesley. ISBN 0-32182662-0.
known today as a dictionary or hash. Dictionaries con• McCreary, Dan; Kelly, Ann (2013). Making Sense
tain a collection of objects, or records, which in turn have
of NoSQL: A guide for managers and the rest of us.
many different fields within them, each containing data.
ISBN 9781617291074.
These records are stored and retrieved using a key that
• Strauch, Christof (2012). “NoSQL Databases” uniquely identifies the record, and is used to quickly find
the data within the database.
Key-value stores work in a very different fashion than
• Moniruzzaman, A. B.; Hossain, S. A. (2013).
the better known relational databases (RDB). RDBs pre“NoSQL Database: New Era of Databases for Big
define the data structure in the database as a series of tadata Analytics - Classification, Characteristics and
bles containing fields with well defined data types. ExComparison”. arXiv:1307.0191.
posing the data types to the database program allows it to
• Orend, Kai (2013). “Analysis and Classification of apply a number of optimizations. In contrast, key-value
NoSQL Databases and Evaluation of their Ability systems treat the data as a single opaque collection which
to Replace an Object-relational Persistence Layer”. may have different fields for every record. This offers
considerable flexibility and more closely follows modern
concepts like object-oriented programming. Because op• Krishnan, Ganesh; Kulkarni, Sarang; Dadbhawala, tional values are not represented by placeholders as in
Dharmesh Kirit. “Method and system for versioned most RDBs, key-value stores often use far less memory
sharing, consolidating and reporting information”.
to store the same database, which can lead to large performance gains in certain workloads.
Performance, a lack of standardization and other issues
limited key-value systems to niche uses for many years,
Strauch, Christoph. “NoSQL whitepaper” (PDF). but the rapid move to cloud computing after 2010 has
led to a renaissance as part of the broader NoSQL moveStuttgart: Hochschule der Medien.
ment. A subclass of the key-value store is the documentoriented database, which offers additional tools that use
Edlich, Stefan. “NoSQL database List”.
the metadata in the data to provide a richer key-value
Neubauer, Peter (2010).
“Graph Databases, database that more closely matches the use patterns of
NOSQL and Neo4j”.
RDBM systems. Some graph databases are also keyvalue stores internally, adding the concept of the relationBushik, Sergey (2012). “A vendor-independent
ships (pointers) between records as a first class data type.
comparison of NoSQL databases: Cassandra,
HBase, MongoDB, Riak”. NetworkWorld.
External links
• Zicari, Roberto V. (2014). “NoSQL Data Stores –
Articles, Papers, Presentations”.
2.6.1 Types and notable examples
Key-value stores can use consistency models ranging
from eventual consistency to serializability. Some support ordering of keys. Some maintain data in memory
2.6 Key-value database
(RAM), while others employ solid-state drives or rotating
A key-value store, or key-value database, is a data storage paradigm designed for storing, retrieving, and man- Redis was the most popular implementation of a keyaging associative arrays, a data structure more commonly value database as of August 2015, according to DB-
Engines Ranking.[1]
• Hazelcast
Another example of key-value database is Oracle NoSQL
• memcached
Database. Oracle NoSQL Database provides a key-value
paradigm to the application developer. Every entity
• OpenLink Virtuoso
(record) is a set of key-value pairs. A key has multiple components, specified as an ordered list. The ma• Redis
jor key identifies the entity and consists of the leading
components of the key. The subsequent components
are called minor keys. This organization is similar to a
• Gemfire
directory path specification in a file system (e.g., /Major/minor1/minor2/). The “value” part of the key-value
pair is simply an uninterpreted string of bytes of arbitrary
KV - solid-state drive or rotating disk
The Unix system provides dbm (DataBase Manager)
which is a library originally written by Ken Thompson.
The dbm manages associative arrays of arbitrary data by
use of a single key (a primary key). Modern implementations include ndbm, sdbm and GNU dbm.
• Aerospike
• BigTable
• Clusterpoint Database Server
KV - eventually consistent
• Apache Cassandra
• Dynamo
• Oracle NoSQL Database
• Project Voldemort
• Riak[3]
• OpenLink Virtuoso
KV - ordered
• Couchbase Server
• FairCom c-treeACE
• GT.M[4]
• Hibari
• Keyspace
• LevelDB
• MemcacheDB (using Berkeley DB or LMDB)
• Berkeley DB
• MongoDB
• FairCom c-treeACE/c-treeRTG
• NoSQLz
• FoundationDB
• Coherence
• HyperDex
• Oracle NoSQL Database
• IBM Informix C-ISAM
• OpenLink Virtuoso
• InfinityDB
• Tarantool
• MemcacheDB
• Tokyo Cabinet
• Tuple space
2.6.2 References
• Aerospike
• Coherence
• FairCom c-treeACE
[2] “Oracle NoSQL Database”
[3] “Riak: An Open Source Scalable Data Store”. 28 November 2010. Retrieved 28 November 2010.
[4] Tweed, Rob; James, George (2010). “A Universal
NoSQL Engine, Using a Tried and Tested Technology”
(PDF). p. 25. Without exception, the most successful and
well-known of the NoSQL databases have been developed
from scratch, all within just the last few years. Strangely,
it seems that nobody looked around to see whether there
were any existing, successfully implemented database
technologies that could have provided a sound foundation for meeting Web-scale demands. Had they done
so, they might have discovered two products, GT.M and
External links
• Ranking of key-value databases by popularity
2.7 Document-oriented database
tinual change in place, and speed of deployment is an important issue.
2.7.1 Documents
The central concept of a document-oriented database are
the documents, which is used in usual English sense of a
group of data that encodes some sort of user-readable information. This contrasts with the value in the key-value
store, which is assumed to be opaque data. The basic
concept that makes a database document-oriented as opposed to key-value is the idea that the documents include
internal structure, or metadata, that the database engine
can use to further automate the storage and provide more
To understand the difference, consider this text document:
This article is about the software type.
For us- Bob Smith 123 Back St. Boys, AR, 32225 US
age/deployment instances, see Full text database.
Although it is clear to the reader that this document contains the address for a contact, there is no information
A document-oriented database or document store is within the document that indicates that, nor information
a computer program designed for storing, retrieving, and on what the individual fields represent. This file could be
managing document-oriented information, also known as stored in a key-value store, but the semantic content that
semi-structured data. Document-oriented databases are this is an address may be lost, and the database has no
one of the main categories of NoSQL databases and the way to know how to optimize or index this data by itself.
popularity of the term “document-oriented database” has For instance, there is no way for the database to know that
“AR” is the state and add it to an index of states, it is simgrown[1] with the use of the term NoSQL itself.
ply a piece of data in a string that also includes the city and
Document-oriented databases are inherently a subclass of zip code. It is possible to add additional logic to deconthe key-value store, another NoSQL database concept. struct the string into fields, to extract the state by looking
The difference lies in the way the data is processed; in for the middle item of three comma separated values in
a key-value store the data is considered to be inherently the 3rd line, but this is not a simple task. For instance, if
opaque to the database, whereas a document-oriented another line is added to the address, adding a PO Box or
system relies on internal structure in the document order suite number for instance, the state information is in the
to extract metadata that the database engine uses for fur- 4th line instead of 3rd. Without additional information,
ther optimization. Although the difference is often moot parsing free form data of this sort can be complex.
due to tools in the systems,[lower-alpha 1] conceptually the
Now consider the same document marked up in pseudodocument-store is designed to offer a richer experience
with modern programming techniques. XML databases XML:
<lastare a specific subclass of document-oriented databases <contact>
that are optimized to extract their metadata from XML name>Smith</lastname>
St.</street1> <city>Boys</city> <state>AR</state>
Document databases[lower-alpha 2] contrast strongly with <zip>32225</zip> <country>US</country> </contact>
the traditional relational database (RDB). Relational
databases are strongly typed during database creation,
and store repeated data in separate tables that are defined
by the programmer. In an RDB, every instance of data
has the same format as every other, and changing that format is generally difficult. Document databases get their
type information from the data itself, normally store all
related information together, and allow every instance of
data to be different from any other. This makes them
more flexible in dealing with change and optional values,
maps more easily into program objects, and often reduces
database size. This makes them attractive for programming modern web applications, which are subject to con-
In this case, the document includes both data and the
metadata explaining each of the fields. A key-value store
receiving this document would simply store it. In the case
of a document-store, the system understands that contact
documents may have a state field, allowing the programmer to “find all the <contact>s where the <state> is 'AR'".
Additionally, the programmer can provide hints based on
the document type or fields within it, for instance, they
may tell the engine to place all <contact> documents in a
separate physical store, or to make an index on the state
field for performance reasons. All of this can be done
in a key-value store as well, and the difference lies pri-
marily in how much programming effort is needed to add used; a document-oriented database is designed from the
these indexes and other features; in a document-store this start to work with complex documents, and will (hopeis normally almost entirely automated.
fully) make it easier to access this functionality than a
system where this was added after the fact.
Now consider a slightly more complex example:
type="Home">[email protected]</email>
<address> <type>Home</type> <street1>123 Back
St.</street1> <city>Boys</city> <state>AR</state>
<zip>32225</zip> <country>US</country> </address>
Practically any “document” containing metadata can be
managed in this fashion, and common examples include XML, YAML, JSON, and BSON. Some documentoriented databases include functionality to help map data
lacking clearly defined metadata. For instance, many engines include functionality to index PDF or TeX documents, or may include predefined document formats
that are in turn based on XML, like MathML, JATS or
DocBook. Some allow documents to be mapped onto
a more suitable format using a schema language such as
DTD, XSD, Relax NG, or Schematron. Others may include tools to map enterprise data, like column-delimited
text files, into formats that can be read more easily by the
database engine. Still others take the opposite route, and
are dedicated to one type of data format, JSON. JSON
is widely used in online programming for interactive web
pages and mobile apps, and a niche has appeared for document stores dedicated to efficiently handling them.
In this case a number of the fields are either repeated or
split out into separate containers in the case of <address>.
With similar hints, the document store will allow searches
for things like “find all my <contact>s with a <phone>
of type <work> but does not have an <email> of type
<work>". This is not unlike other database systems in
terms of retrieval. What is different is that these fields
are defined by the metadata in the document itself. There
Some of the most popular Web sites are document
is no need to pre-define these fields in the database.
databases, including the many collections of articles at
This is another major advantage of the document- or major journal publishers; Wikipedia and
oriented concept; a single database can contain both of its kin; and even search engines (though many of those
these <contact> objects in the same store, and more gen- store links to indexed documents, rather than the full docerally, every document in the database can have a differ- uments themselves).
ent format. It is very common for a particular type of document to differ from instance to instance; one <contact>
might have a work email, another might not, one might Keys and retrieval
have a single address, another might have several. More
widely, the database can store completely unrelated doc- Documents may be addressed in the database via a unique
uments, yet still understand that parts of the data within key that represents that document. This key is often
them are the same. For instance, one could construct a simple string, a URI, or a path. The key can be
a query that would look for any document that has the used to retrieve the document from the database. Typi<state> 'AR', it doesn't matter that the documents might cally, the database retains an index on the key to speed
be <contact>s or <business>es, or if the <state> is within up document retrieval. The most primitive document
databases may do little more than that. However, modern
an <address> or not.
document-oriented databases provide far more, because
In addition to making it easier to handle different types
they extract and index all kinds of metadata, and usuof data, the metadata also allows the document format
ally also the entire data content, of the documents. Such
to be changed at any time without affecting the existdatabases offer a query language that allows the user to
ing records. If one wishes to add an <image> field to
retrieve documents based on their content. For example,
their contact book application some time in the future,
you may want to retrieve all the documents whose date
they simply add it. Existing documents will still work
falls within some range, that contains a citation to another
fine without being changed in the database, they simply
document, etc.. The set of query APIs or query language
won't have an image. Fields can be added at any time,
features available, as well as the expected performance of
anywhere, with no need to change the physical storage.
the queries, varies significantly from one implementation
The usefulness of this sort of introspection of the data to the next.
is not lost on the designers of other database systems.
Many key-value stores include some or all of the functionality of dedicated from the start document stores, and a Organization
number of relational databases, notably PostgreSQL and
Informix, have added functionality to make these sorts of Implementations offer a variety of ways of organizing
operations possible. It is not the ability to provide these documents, including notions of:
functions that define the document-orientation, but the
ease with which these functions can be implemented and
• Collections
• Tags
• Non-visible Metadata
• Directory hierarchies
• Buckets
can change in type and form at any time. If one wishes to
add a COUNTRY_FLAG to a CONTACT, simply add
this field to new documents as they are inserted, this will
have no effect on the database or the existing documents
already stored, they simply won't have this field. This indicates an advantage of the document-based model: optional fields are truly optional, so a contact that does not
include a mailing address simply does not have a mailing
relational address, and there is no need to check another table to see
if there are entries.
To aid retrieval of information from the database,
document-oriented systems generally allow the administrator to provide hints to the database to look for certain
types of information. In the address book example, the
design might add hints for the first and last name fields.
When the document is inserted into the database (or later
modified), the database engine looks for these bits of information and indexes them, in the same fashion as the
relational model. Additionally, most document-oriented
databases allow documents to have a type associated with
them, like “address book entry”, which allows the programmer to retrieve related types of information, like “all
the address book entries”. This provides functionality
For example, an address book application will gener- similar to a table, but separates the concept (categories
ally need to store the contact name, an optional im- of data) from its physical implementation (tables).
age, one or more phone numbers, one or more mail- All of this is predicated on the ability of the database
ing addresses, and one or more email addresses. In engine to examine the data in the document and extract
a canonical relational database solution, tables would fields from the formatting, its metadata. This is easy in the
be created for each of these records with prede- case of, for example, an XML document or HTML page,
fined fields for each bit of data: the CONTACT where markup tags clearly identify various bits of data.
table might include FIRST_NAME, LAST_NAME Document-oriented databases may include functionality
and IMAGE fields, while the PHONE_NUMBER ta- to automatically extract this sort of information from a
ble might include COUNTRY_CODE, AREA_CODE, variety of document types, even those that were not origPHONE_NUMBER and TYPE (home, work, etc). The inally designed for easy access in this manner. In other
PHONE_NUMBER table also contains a foreign key cases the programmer has to provide this information usfield, “CONTACT_ID”, which holds the unique ID num- ing their own code. In contrast, a relational database reber assigned to the contact when it was created. In order lies on the programmer to handle all of these tasks, breakto recreate the original contact, the system has to search ing down the document into fields and providing those to
through all of the tables and collect the information back the database engine, which may require separate instructogether using joins.
tions if the data spans tables.
In a relational database, data is first categorized into a
number of predefined types, and tables are created to hold
individual entries, or records, of each type. The tables
define the data within each record’s fields, meaning that
every record in the table has the same overall form. The
administrator also defines the relations between the tables, and selects certain fields that they believe will be
most commonly used for searching and defines indexes
on them. A key concept in the relational design is that
any data that may be repeated is placed in its own table,
and if these instances are related to each other, a field is
selected to group them together, the foreign key.
In contrast, in a document-oriented database there may
be no internal structure that maps directly onto the concept of a table, and the fields and relations generally don't
exist as predefined concepts. Instead, all of the data for
an object is placed in a single document, and stored in the
database as a single entry. In the address book example,
the document would contain the contact’s name, image,
and any contact info, all in a single record. That entry is
accessed through a key, some unique bit of data, which
allows the database to retrieve and return the document
to the application. No additional work is needed to retrieve the related data; all of this is returned in a single
Document-oriented databases normally map more
cleanly onto existing programming concepts, like
object-oriented programming (OOP). OOP systems
have a structure somewhere between the relational and
document models; they have predefined fields but they
may be empty, they have a defined structure but that may
change, they have related data store in other objects,
but they may be optional, and collections of other data
are directly linked to the “master” object; there is no
need to look in other collections to gather up related
information. Generally, any object that can be archived
to a document can be stored directly in the database and
directly retrieved. Most modern OOP systems include
A key difference between the document-oriented and re- archiving systems as a basic feature.
lational models is that the data formats are not predefined The relational model stores each part of the object as
in the document case. In most cases, any sort of docu- a separate concept and has to split out this information
ment can be stored in any database, and those documents
on storage and recombine it on retrieval. This leads to
a problem known as object-relational impedance mismatch, which requires considerable effort to overcome.
Object-relational mapping systems, which solve these
problems, are often complex and have a considerable performance overhead. This problem simply doesn't exist
in a document-oriented system, and more generally, in
NoSQL systems as a whole.
[5] CouchDB Document API
[7] eXist-db Open Source Native XML Database. Retrieved on 2013-09-18.
Main category: Document-oriented databases
[10] MongoDB Licensing
[11] Additional 30+ community MongoDB supported drivers
[12] MongoDB REST Interfaces
XML database implementations
Further information: XML database
Most XML databases are document-oriented databases.
[13] GTM MUMPS FOSS on SourceForge
2.7.7 Further reading
• Assaf Arkin. (2007, September 20). Read Consistency: Dumb Databases, Smart Services. Labnotes:
Don’t let the bubble go to your head!
See also
• Database theory
• Data hierarchy
• Full text search
• In-memory database
• Internet Message Access Protocol (IMAP)
• Object database
• Online database
• Real time database
• Relational database
2.7.8 External links
• DB-Engines Ranking of Document Stores by popularity, updated monthly
2.8 NewSQL
NewSQL is a class of modern relational database management systems that seek to provide the same scalable performance of NoSQL systems for online transaction processing (OLTP) read-write workloads while
still maintaining the ACID guarantees of a traditional
database system.[1][2][3]
2.8.1 History
The term was first used by 451 Group analyst Matthew
Aslett in a 2011 research paper discussing the rise of new
[1] To the point that document-oriented and key-value sys- database systems as challengers to established vendors.[4]
tems can often be interchanged in operation.
Many enterprise systems that handle high-profile data
(e.g., financial and order processing systems) also need
[2] And key-value stores in general.
to be able to scale but are unable to use NoSQL solutions
because they cannot give up strong transactional and consistency requirements.[4][5] The only options previously
2.7.6 References
available for these organizations were to either purchase
a more powerful single-node machine or develop cus[1] DB-Engines Ranking per database model category
tom middleware that distributes queries over traditional
[2] Document-oriented Database. Clusterpoint. Retrieved on DBMS nodes. Both approaches are prohibitively expen2015-10-08.
sive and thus are not an option for many. Thus, in this paper, Aslett discusses how NewSQL upstarts are poised to
[3] Documentation. Couchbase. Retrieved on 2013-09-18.
challenge the supremacy of commercial vendors, in par[4] CouchDB Overview
ticular Oracle.
Although NewSQL systems vary greatly in their internal architectures, the two distinguishing features common
amongst them is that they all support the relational data
model and use SQL as their primary interface.[6] The applications targeted by these NewSQL systems are characterized as having a large number of transactions that (1)
are short-lived (i.e., no user stalls), (2) touch a small subset of data using index lookups (i.e., no full table scans
or large distributed joins), and (3) are repetitive (i.e. executing the same queries with different inputs).[7] These
NewSQL systems achieve high performance and scalability by eschewing much of the legacy architecture of
the original IBM System R design, such as heavyweight
recovery or concurrency control algorithms.[8] One of
the first known NewSQL systems is the H-Store parallel
database system.[9][10]
NewSQL systems can be loosely grouped into three categories: [11][12]
New architectures
The first type of NewSQL systems are completely new
database platforms. These are designed to operate in a
distributed cluster of shared-nothing nodes, in which each
node owns a subset of the data. These databases are often written from scratch with a distributed architecture in
mind, and include components such as distributed concurrency control, flow control, and distributed query processing. Example systems in this category are Google
Spanner, Clustrix, VoltDB, MemSQL, Pivotal's GemFire
XD, SAP HANA,[13] NuoDB, and Trafodion.[14]
SQL engines
2.8.4 References
[1] Aslett, Matthew (2011). “How Will The Database Incumbents Respond To NoSQL And NewSQL?" (PDF). 451
Group (published 2011-04-04). Retrieved 2012-07-06.
[2] Stonebraker, Michael (2011-06-16). “NewSQL: An Alternative to NoSQL and Old SQL for New OLTP Apps”.
Communications of the ACM Blog. Retrieved 2012-0706.
[3] Hoff, Todd (2012-09-24). “Google Spanner’s Most Surprising Revelation: NoSQL is Out and NewSQL is In”.
Retrieved 2012-10-07.
[4] Aslett, Matthew (2010). “What we talk about when we
talk about NewSQL”. 451 Group (published 2011-0406). Retrieved 2012-10-07.
[5] Lloyd, Alex (2012). “Building Spanner”. Berlin Buzzwords (published 2012-06-05). Retrieved 2012-10-07.
[6] Cattell, R. (2011). “Scalable SQL and NoSQL data
stores” (PDF). ACM SIGMOD Record 39 (4): 12.
[7] Stonebraker, Mike; et al. (2007). “The end of an architectural era: (it’s time for a complete rewrite” (PDF). VLDB
'07: Proceedings of the 33rd international conference on
Very large data bases. Vienna, Austria.
[8] Stonebraker, M.; Cattell, R. (2011).
“10 rules
for scalable performance in 'simple operation' datastores”. Communications of the ACM 54 (6): 72.
[9] Aslett, Matthew (2008). “Is H-Store the future of
database management systems?" (published 2008-03-04).
Retrieved 2012-07-05.
[10] Dignan, Larry (2008). “H-Store: Complete destruction of
the old DBMS order?". Retrieved 2012-07-05.
[11] Venkatesh, Prasanna (2012). “NewSQL - The New Way
to Handle Big Data” (published 2012-01-30). Retrieved
The second category are highly optimized storage engines for SQL. These systems provide the same programming interface as SQL, but scale better than built-in en- [12] Levari, Doron (2011). “The NewSQL Market Breakgines, such as InnoDB. Examples of these new storage
down”. Retrieved 2012-04-08.
engines include MySQL Cluster, Infobright, TokuDB and
[13] “SAP HANA”. SAP. Retrieved 17 July 2014.
the now defunct InfiniDB.
[14] “Trafodion: Transactional SQL-on-HBase”. 2014.
Transparent sharding
These systems provide a sharding middleware layer to automatically split databases across multiple nodes. Examples of this type of system includes dbShards and
See also
• Transaction processing
• Partition (database)
Chapter 3
3.1 ACID
bring the database from one valid state to another. Any
data written to the database must be valid according to
all defined rules, including constraints, cascades, triggers,
For other uses, see Acid (disambiguation).
and any combination thereof. This does not guarantee
correctness of the transaction in all ways the application
In computer science, ACID (Atomicity, Consistency, programmer might have wanted (that is the responsibility
Isolation, Durability) is a set of properties that guaran- of application-level code) but merely that any programtee that database transactions are processed reliably. In ming errors cannot result in the violation of any defined
the context of databases, a single logical operation on the rules.
data is called a transaction. For example, a transfer of
funds from one bank account to another, even involving
multiple changes such as debiting one account and cred- Isolation
iting another, is a single transaction.
Main article: Isolation (database systems)
Jim Gray defined these properties of a reliable transaction
system in the late 1970s and developed technologies to
The isolation property ensures that the concurrent execuachieve them automatically.[1][2][3]
tion of transactions results in a system state that would be
In 1983, Andreas Reuter and Theo Härder coined the
obtained if transactions were executed serially, i.e., one
acronym ACID to describe them.[4]
after the other. Providing isolation is the main goal of
concurrency control. Depending on concurrency control
method (i.e. if it uses strict - as opposed to relaxed - seri3.1.1 Characteristics
alizability), the effects of an incomplete transaction might
not even be visible to another transaction.
The characteristics of these four properties as defined by
Reuter and Härder:
Main article: Durability (database systems)
Main article: Atomicity (database systems)
Atomicity requires that each transaction be “all or nothing": if one part of the transaction fails, the entire transaction fails, and the database state is left unchanged.
An atomic system must guarantee atomicity in each and
every situation, including power failures, errors, and
crashes. To the outside world, a committed transaction
appears (by its effects on the database) to be indivisible
(“atomic”), and an aborted transaction does not happen.
The durability property ensures that once a transaction
has been committed, it will remain so, even in the event of
power loss, crashes, or errors. In a relational database, for
instance, once a group of SQL statements execute, the results need to be stored permanently (even if the database
crashes immediately thereafter). To defend against power
loss, transactions (or their effects) must be recorded in a
non-volatile memory.
3.1.2 Examples
The following examples further illustrate the ACID properties. In these examples, the database table has two
Main article: Consistency (database systems)
columns, A and B. An integrity constraint requires that
the value in A and the value in B must sum to 100. The
The consistency property ensures that any transaction will following SQL code creates a table as described above:
3.1. ACID
CREATE TABLE acidtest (A INTEGER, B INTEGER, If these operations are performed in order, isolation is
CHECK (A + B = 100));
maintained, although T2 must wait. Consider what happens if T1 fails half-way through. The database eliminates T1 's effects, and T2 sees only valid data.
Atomicity failure
By interleaving the transactions, the actual order of actions might be:
In database systems, atomicity (or atomicness; from
• T1 subtracts 10 from A.
Greek a-tomos, undividable) is one of the ACID transaction properties. In an atomic transaction, a series of
• T2 subtracts 10 from B.
database operations either all occur, or nothing occurs.
The series of operations cannot be divided apart and exe• T2 adds 10 to A.
cuted partially from each other, which makes the series of
• T1 adds 10 to B.
operations “indivisible”, hence the name. A guarantee of
atomicity prevents updates to the database occurring only
partially, which can cause greater problems than reject- Again, consider what happens if T1 fails halfway through.
ing the whole series outright. In other words, atomicity By the time T1 fails, T2 has already modified A; it cannot
be restored to the value it had before T1 without leaving
means indivisibility and irreducibility.
an invalid database. This is known as a write-write failure, because two transactions attempted to write to the
Consistency failure
same data field. In a typical system, the problem would
be resolved by reverting to the last known good state, canConsistency is a very general term, which demands that celing the failed transaction T1 , and restarting the interthe data must meet all validation rules. In the previous ex- rupted transaction T2 from the good state.
ample, the validation is a requirement that A + B = 100.
Also, it may be inferred that both A and B must be integers. A valid range for A and B may also be inferred. All Durability failure
validation rules must be checked to ensure consistency.
Assume that a transaction attempts to subtract 10 from A Consider a transaction that transfers 10 from A to B. First
without altering B. Because consistency is checked after it removes 10 from A, then it adds 10 to B. At this point,
each transaction, it is known that A + B = 100 before the the user is told the transaction was a success, however the
transaction begins. If the transaction removes 10 from A changes are still queued in the disk buffer waiting to be
successfully, atomicity will be achieved. However, a vali- committed to disk. Power fails and the changes are lost.
dation check will show that A + B = 90, which is inconsis- The user assumes (understandably) that the changes have
tent with the rules of the database. The entire transaction been persisted.
must be cancelled and the affected rows rolled back to
their pre-transaction state. If there had been other con- 3.1.3 Implementation
straints, triggers, or cascades, every single change operation would have been checked in the same way as above Processing a transaction often requires a sequence of opbefore the transaction was committed.
erations that is subject to failure for a number of reaIsolation failure
To demonstrate isolation, we assume two transactions execute at the same time, each attempting to modify the
same data. One of the two must wait until the other completes in order to maintain isolation.
Consider two transactions. T1 transfers 10 from A to B.
T2 transfers 10 from B to A. Combined, there are four
• T1 subtracts 10 from A.
• T1 adds 10 to B.
• T2 subtracts 10 from B.
• T2 adds 10 to A.
sons. For instance, the system may have no room left
on its disk drives, or it may have used up its allocated
CPU time. There are two popular families of techniques:
write-ahead logging and shadow paging. In both cases,
locks must be acquired on all information to be updated,
and depending on the level of isolation, possibly on all
data that be read as well. In write ahead logging, atomicity is guaranteed by copying the original (unchanged)
data to a log before changing the database. That allows
the database to return to a consistent state in the event of
a crash. In shadowing, updates are applied to a partial
copy of the database, and the new copy is activated when
the transaction commits.
Locking vs multiversioning
Many databases rely upon locking to provide ACID capabilities. Locking means that the transaction marks the
data that it accesses so that the DBMS knows not to al• Transactional NTFS
low other transactions to modify it until the first transac•
tion succeeds or fails. The lock must always be acquired
before processing data, including data that is read but
not modified. Non-trivial transactions typically require
3.1.5 References
a large number of locks, resulting in substantial overhead
as well as blocking other transactions. For example, if [1] “Gray to be Honored With A.M. Turing Award This
user A is running a transaction that has to read a row of
Spring”. Microsoft PressPass. Archived from the origidata that user B wants to modify, user B must wait unnal on February 6, 2009. Retrieved March 27, 2015.
til user A’s transaction completes. Two phase locking is
[2] Gray, Jim (September 1981). “The Transaction Concept:
often applied to guarantee full isolation.
An alternative to locking is multiversion concurrency control, in which the database provides each reading transaction the prior, unmodified version of data that is being
modified by another active transaction. This allows readers to operate without acquiring locks, i.e. writing transactions do not block reading transactions, and readers do
not block writers. Going back to the example, when user
A’s transaction requests data that user B is modifying, the
database provides A with the version of that data that existed when user B started his transaction. User A gets a
consistent view of the database even if other users are
changing data. One implementation, namely snapshot
isolation, relaxes the isolation property.
Distributed transactions
Virtues and Limitations” (PDF). Proceedings of the 7th International Conference on Very Large Databases. Cupertino, CA: Tandem Computers. pp. 144–154. Retrieved
March 27, 2015.
[3] Gray, Jim & Andreas Reuter. Distributed Transaction
Processing: Concepts and Techniques. Morgan Kaufmann,
1993; ISBN 1-55860-190-2.
[4] Haerder, T.; Reuter, A. (1983).
“Principles of
transaction-oriented database recovery”. ACM Computing
Surveys 15 (4): 287. doi:10.1145/289.291. These four
properties, atomicity, consistency, isolation, and durability (ACID), describe the major highlights of the transaction paradigm, which has influenced many aspects of development in database systems.
3.2 Consistency (database systems)
Main article: Distributed transaction
Consistency in database systems refers to the requirement that any given database transaction must change affected data only in allowed ways. Any data written to the
database must be valid according to all defined rules, including constraints, cascades, triggers, and any combination thereof. This does not guarantee correctness of the
transaction in all ways the application programmer might
have wanted (that is the responsibility of application-level
code) but merely that any programming errors cannot result in the violation of any defined rules.
Guaranteeing ACID properties in a distributed transaction across a distributed database, where no single node
is responsible for all data affecting a transaction, presents
additional complications. Network connections might
fail, or one node might successfully complete its part
of the transaction and then be required to roll back its
changes because of a failure on another node. The twophase commit protocol (not to be confused with twophase locking) provides atomicity for distributed transactions to ensure that each participant in the transaction
agrees on whether the transaction should be committed
3.2.1 As an ACID guarantee
or not. Briefly, in the first phase, one node (the coordinator) interrogates the other nodes (the participants) and
Consistency is one of the four guarantees that define
only when all reply that they are prepared does the coorACID transactions; however, significant ambiguity exists
dinator, in the second phase, formalize the transaction.
about the nature of this guarantee. It is defined variously
See also
• Basically Available, Soft state, Eventual consistency
• CAP theorem
• Concurrency control
• Java Transaction API
• Open Systems Interconnection
• The guarantee that any transactions started in the future necessarily see the effects of other transactions
committed in the past[1][2]
• The guarantee that database constraints are
not violated, particularly once a transaction
• The guarantee that operations in transactions are
performed accurately, correctly, and with validity,
with respect to application semantics[7]
As these various definitions are not mutually exclusive,
it is possible to design a system that guarantees “consistency” in every sense of the word, as most relational
database management systems in common use today arguably do.
As a CAP trade-off
See also
will survive permanently. For example, if a flight booking reports that a seat has successfully been booked, then
the seat will remain booked even if the system crashes.
Durability can be achieved by flushing the transaction’s
log records to non-volatile storage before acknowledging
In distributed transactions, all participating servers must
coordinate before commit can be acknowledged. This is
The CAP Theorem is based on three trade-offs, one of usually done by a two-phase commit protocol.
which is “atomic consistency” (shortened to “consistency” Many DBMSs implement durability by writing transacfor the acronym), about which the authors note, “Dis- tions into a transaction log that can be reprocessed to
cussing atomic consistency is somewhat different than recreate the system state right before any later failure. A
talking about an ACID database, as database consistency transaction is deemed committed only after it is entered
refers to transactions, while atomic consistency refers in the log.
only to a property of a single request/response operation
sequence. And it has a different meaning than the Atomic
in ACID, as it subsumes the database notions of both 3.3.1 See also
Atomic and Consistent.”[1]
• Atomicity
• Consistency model
• CAP Theorem
• Eventual consistency
brewer_proof.pdf “Brewer’s Conjecture and the Feasibility of Consistent, Available, Partition-Tolerant Web
[2] Ports, D.R.K, Clements, A.T, Zhang, I, Madden, S,
Liskov, B. “Transactional Consistency and Automatic
Management in an Application Data Cache” (PDF). MIT
[3] Haerder, T, Reuter, A. (December 1983). “Principles of
Transaction-Oriented Database Recovery” (PDF). Computing Surveys 15 (4): 287–317.
[4] Mike Chapple. “The ACID Model”. About.
[5] “ACID properties”.
[6] Cory Janssen. “What is ACID in Databases? - Definition
from Techopedia”.
[7] “ISO/IEC 10026-1:1998 - Information technology -Open Systems Interconnection -- Distributed Transaction
Processing -- Part 1: OSI TP Model”.
3.3 Durability (database systems)
In database systems, durability is the ACID property
which guarantees that transactions that have committed
• Consistency
• Isolation
• Relational database management system
Chapter 4
4.1 Serializability
distinguish from a transaction request, waiting to get execution resources) can be in one of three states, or phases:
In concurrency control of databases,[1][2] transaction
1. Running - Its program(s) is (are) executing.
processing (transaction management), and various
transactional applications (e.g., transactional memory
2. Ready - Its program’s execution has ended, and it is
and software transactional memory), both centralized
waiting to be Ended (Completed).
and distributed, a transaction schedule is serializable if
its outcome (e.g., the resulting database state) is equal
3. Ended (or Completed) - It is either Committed or
to the outcome of its transactions executed serially, i.e.,
Aborted (Rolled-back), depending whether the exsequentially without overlapping in time. Transactions
ecution is considered a success or not, respectively .
are normally executed concurrently (they overlap), since
When committed, all its recoverable (i.e., with states
this is the most efficient way. Serializability is the
that can be controlled for this purpose), durable remajor correctness criterion for concurrent transactions’
sources (typically database data) are put in their fiexecutions. It is considered the highest level of isolation
nal states, states after running. When aborted, all
between transactions, and plays an essential role in
its recoverable resources are put back in their initial
concurrency control. As such it is supported in all genstates, as before running.
eral purpose database systems. Strong strict two-phase
locking (SS2PL) is a popular serializability mechanism
utilized in most of the database systems (in various A failure in transaction’s computing environment before
ending typically results in its abort. However, a transacvariants) since their early days in the 1970s.
tion may be aborted also for other reasons as well (e.g.,
Serializability theory provides the formal framework see below).
to reason about and analyze serializability and its techniques. Though it is mathematical in nature, its funda- Upon being ended (completed), transaction’s allocated
mentals are informally (without mathematics notation) computing resources are released and the transaction disappears from the computing environment. However, the
introduced below.
effects of a committed transaction remain in the database,
while the effects of an aborted (rolled-back) transaction
disappear from the database. The concept of atomic
4.1.1 Database transaction
transaction (“all or nothing” semantics) was designed to
exactly achieve this behavior, in order to control correctMain article: Database transaction
ness in complex faulty systems.
For this is a specific intended run (with specific parameters, e.g., with transaction identification, at least) of a
computer program (or programs) that accesses a database
(or databases). Such a program is written with the assumption that it is running in isolation from other executing programs, i.e., when running, its accessed data
(after the access) are not changed by other running programs. Without this assumption the transaction’s results
are unpredictable and can be wrong. The same transaction can be executed in different situations, e.g., in different times and locations, in parallel with different programs. A live transaction (i.e., exists in a computing environment with already allocated computing resources; to
4.1.2 Correctness
Correctness - serializability
Serializability is a property of a transaction schedule
(history). It relates to the isolation property of a database
Serializability of a schedule means equivalence (in the outcome, the database state, data
values) to a serial schedule (i.e., sequential with
no transaction overlap in time) with the same
transactions. It is the major criterion for the
correctness of concurrent transactions’ schedule, and thus supported in all general purpose
database systems.
The rationale behind serializability is the
If each transaction is correct by itself, i.e.,
meets certain integrity conditions, then a
schedule that comprises any serial execution
of these transactions is correct (its transactions
still meet their conditions): “Serial” means that
transactions do not overlap in time and cannot
interfere with each other, i.e, complete isolation between each other exists. Any order of
the transactions is legitimate, if no dependencies among them exist, which is assumed (see
comment below). As a result, a schedule that
comprises any execution (not necessarily serial) that is equivalent (in its outcome) to any
serial execution of these transactions, is correct.
correctness is not harmed), compromising recoverability
would quickly violate the database’s integrity, as well as
that of transactions’ results external to the database. A
schedule with the recoverability property (a recoverable
schedule) “recovers” from aborts by itself, i.e., aborts do
not harm the integrity of its committed transactions and
resulting database. This is false without recoverability,
where the likely integrity violations (resulting incorrect
database data) need special, typically manual, corrective
actions in the database.
Implementing recoverability in its general form may result in cascading aborts: Aborting one transaction may
result in a need to abort a second transaction, and then a
third, and so on. This results in a waste of already partially executed transactions, and may result also in a performance penalty. Avoiding cascading aborts (ACA, or
Cascadelessness) is a special case of recoverability that
exactly prevents such phenomenon. Often in practice a
special case of ACA is utilized: Strictness. Strictness
allows an efficient database recovery from failure.
Note that the recoverability property is needed even if
no database failure occurs and no database recovery from
failure is needed. It is rather needed to correctly automatSchedules that are not serializable are likely to gener- ically handle aborts, which may be unrelated to database
ate erroneous outcomes. Well known examples are with failure and recovery from failure.
transactions that debit and credit accounts with money:
If the related schedules are not serializable, then the total sum of money may not be preserved. Money could Relaxing serializability
disappear, or be generated from nowhere. This and violations of possibly needed other invariant preservations In many applications, unlike with finances, absolute corare caused by one transaction writing, and “stepping on” rectness is not needed. For example, when retrieving a
and erasing what has been written by another transaction list of products according to specification, in most cases
before it has become permanent in the database. It does it does not matter much if a product, whose data was upnot happen if serializability is maintained.
dated a short time ago, does not appear in the list, even
if it meets the specification. It will typically appear in
such a list when tried again a short time later. Commercial databases provide concurrency control with a whole
range of isolation levels which are in fact (controlled) serializability violations in order to achieve higher performance. Higher performance means better transaction execution rate and shorter average transaction response time
(transaction duration). Snapshot isolation is an example
of a popular, widely utilized efficient relaxed serializabilA major characteristic of a database transaction is ity method with many characteristics of full serializabilatomicity, which means that it either commits, i.e., all its ity, but still short of some, and unfit in many situations.
operations’ results take effect in the database, or aborts Another common reason nowadays for distributed se(rolled-back), all its operations’ results do not have any rializability relaxation (see below) is the requirement
effect on the database (“all or nothing” semantics of a of availability of internet products and services. This
transaction). In all real systems transactions can abort requirement is typically answered by large-scale data
for many reasons, and serializability by itself is not suf- replication. The straightforward solution for synchronizficient for correctness. Schedules also need to possess ing replicas’ updates of a same database object is includthe recoverability (from abort) property. Recoverabil- ing all these updates in a single atomic distributed transity means that committed transactions have not read data action. However, with many replicas such a transaction is
written by aborted transactions (whose effects do not ex- very large, and may span several computers and networks
ist in the resulting database states). While serializability is that some of them are likely to be unavailable. Thus
currently compromised on purpose in many applications such a transaction is likely to end with abort and miss its
for better performance (only in cases when application’s purpose.[4] Consequently, Optimistic replication (Lazy
If any specific order between some transactions is requested by an application, then it is enforced independently of the underlying serializability mechanisms.
These mechanisms are typically indifferent to any specific order, and generate some unpredictable partial order that is typically compatible with multiple serial orders
of these transactions. This partial order results from the
scheduling orders of concurrent transactions’ data access
operations, which depend on many factors.
replication) is often utilized (e.g., in many products and
services by Google, Amazon, Yahoo, and alike), while
serializability is relaxed and compromised for eventual
consistency. Again in this case, relaxation is done only
for applications that are not expected to be harmed by
this technique.
Classes of schedules defined by relaxed serializability
properties either contain the serializability class, or are
incomparable with it.
View and conflict serializability
Mechanisms that enforce serializability need to execute
in real time, or almost in real time, while transactions are
running at high rates. In order to meet this requirement
special cases of serializability, sufficient conditions for
serializability which can be enforced effectively, are utilized.
Two major types of serializability exist:
viewserializability, and conflict-serializability.
Viewserializability matches the general definition of
serializability given above. Conflict-serializability is
a broad special case, i.e., any schedule that is conflictserializable is also view-serializable, but not necessarily
the opposite.
Conflict-serializability is widely utilized because it is easier to determine and covers a
substantial portion of the view-serializable schedules.
Determining view-serializability of a schedule is an
NP-complete problem (a class of problems with only
difficult-to-compute, excessively time-consuming known
View-serializability of a schedule is defined
by equivalence to a serial schedule (no overlapping transactions) with the same transactions, such that respective transactions in the
two schedules read and write the same data values (“view” the same data values).
Conflict-serializability is defined by equivalence to a serial schedule (no overlapping transactions) with the same transactions, such that
both schedules have the same sets of respective
chronologically ordered pairs of conflicting operations (same precedence relations of respective conflicting operations).
for complex operations, which may consist each of several “simple” read/write operations) requires that they are
noncommutative (changing their order also changes their
combined result). Each such operation needs to be atomic
by itself (by proper system support) in order to be considered an operation for a commutativity check. For example, read-read operations are commutative (unlike readwrite and the other possibilities) and thus read-read is not
a conflict. Another more complex example: the operations increment and decrement of a counter are both write
operations (both modify the counter), but do not need to
be considered conflicting (write-write conflict type) since
they are commutative (thus increment-decrement is not
a conflict; e.g., already has been supported in the old
IBM’s IMS “fast path”). Only precedence (time order) in
pairs of conflicting (non-commutative) operations is important when checking equivalence to a serial schedule,
since different schedules consisting of the same transactions can be transformed from one to another by changing orders between different transactions’ operations (different transactions’ interleaving), and since changing orders of commutative operations (non-conflicting) does
not change an overall operation sequence result, i.e., a
schedule outcome (the outcome is preserved through order change between non-conflicting operations, but typically not when conflicting operations change order). This
means that if a schedule can be transformed to any serial schedule without changing orders of conflicting operations (but changing orders of non-conflicting, while preserving operation order inside each transaction), then the
outcome of both schedules is the same, and the schedule
is conflict-serializable by definition.
Conflicts are the reason for blocking transactions and delays (non-materialized conflicts), or for aborting transactions due to serializability violations prevention. Both
possibilities reduce performance. Thus reducing the
number of conflicts, e.g., by commutativity (when possible), is a way to increase performance.
A transaction can issue/request a conflicting operation
and be in conflict with another transaction while its
conflicting operation is delayed and not executed (e.g.,
blocked by a lock). Only executed (materialized) conflicting operations are relevant to conflict serializability (see
more below).
4.1.4 Enforcing conflict serializability
Testing conflict serializability
Operations upon data are read or write (a write: either insert or modify or delete). Two operations are conflicting,
if they are of different transactions, upon the same datum
(data item), and at least one of them is write. Each such
pair of conflicting operations has a conflict type: It is either a read-write, or write-read, or a write-write conflict.
The transaction of the second operation in the pair is said
to be in conflict with the transaction of the first operation.
A more general definition of conflicting operations (also
Schedule compliance with conflict serializability can be
tested with the precedence graph (serializability graph,
serialization graph, conflict graph) for committed transactions of the schedule. It is the directed graph representing
precedence of transactions in the schedule, as reflected by
precedence of conflicting operations in the transactions.
In the precedence graph transactions are
nodes and precedence relations are directed
edges. There exists an edge from a first transaction to a second transaction, if the second
transaction is in conflict with the first (see Conflict serializability above), and the conflict is
materialized (i.e., if the requested conflicting operation is actually executed: in many
cases a requested/issued conflicting operation
by a transaction is delayed and even never executed, typically by a lock on the operation’s
object, held by another transaction, or when
writing to a transaction’s temporary private
workspace and materializing, copying to the
database itself, upon commit; as long as a requested/issued conflicting operation is not executed upon the database itself, the conflict is
non-materialized; non-materialized conflicts
are not represented by an edge in the precedence graph).
Comment: In many text books only committed transactions are included in the precedence
graph. Here all transactions are included for
convenience in later discussions.
Common mechanism - SS2PL
Main article: Two-phase locking
Strong strict two phase locking (SS2PL) is a common
mechanism utilized in database systems since their early
days in the 1970s (the “SS” in the name SS2PL is
newer though) to enforce both conflict serializability and
strictness (a special case of recoverability which allows
effective database recovery from failure) of a schedule.
In this mechanism each datum is locked by a transaction
before accessing it (any read or write operation): The
item is marked by, associated with a lock of a certain
type, depending on operation (and the specific implementation; various models with different lock types exist; in
some models locks may change type during the transaction’s life). As a result, access by another transaction may
be blocked, typically upon a conflict (the lock delays or
completely prevents the conflict from being materialized
and be reflected in the precedence graph by blocking the
conflicting operation), depending on lock type and the
other transaction’s access operation type. Employing an
SS2PL mechanism means that all locks on data on behalf
of a transaction are released only after the transaction has
ended (either committed or aborted).
SS2PL is the name of the resulting schedule property as
The following observation is a key characterization of well, which is also called rigorousness. SS2PL is a special
conflict serializability:
case (proper subset) of Two-phase locking (2PL)
A schedule is conflict-serializable if and only if
its precedence graph of committed transactions
(when only committed transactions are considered) is acyclic. This means that a cycle consisting of committed transactions only is generated in the (general) precedence graph, if and
only if conflict-serializability is violated.
Mutual blocking between transactions results in a deadlock, where execution of these transactions is stalled, and
no completion can be reached. Thus deadlocks need to
be resolved to complete these transactions’ execution and
release related computing resources. A deadlock is a reflection of a potential cycle in the precedence graph, that
would occur without the blocking when conflicts are materialized. A deadlock is resolved by aborting a transaction involved with such potential cycle, and breaking the
cycle. It is often detected using a wait-for graph (a graph
of conflicts blocked by locks from being materialized;
it can be also defined as the graph of non-materialized
conflicts; conflicts not materialized are not reflected in
the precedence graph and do not affect serializability),
which indicates which transaction is “waiting for” lock
release by which transaction, and a cycle means a deadlock. Aborting one transaction per cycle is sufficient to
break the cycle. Transactions aborted due to deadlock
resolution are restarted and executed again immediately.
Cycles of committed transactions can be prevented by
aborting an undecided (neither committed, nor aborted)
transaction on each cycle in the precedence graph of all
the transactions, which can otherwise turn into a cycle
of committed transactions (and a committed transaction
cannot be aborted). One transaction aborted per cycle is
both required and sufficient number to break and eliminate the cycle (more aborts are possible, and can happen
in some mechanisms, but unnecessary for serializability).
The probability of cycle generation is typically low, but
nevertheless, such a situation is carefully handled, typically with a considerable overhead, since correctness is Other enforcing techniques
involved. Transactions aborted due to serializability violation prevention are restarted and executed again imme- Other known mechanisms include:
Serializability enforcing mechanisms typically do not
maintain a precedence graph as a data structure, but
rather prevent or break cycles implicitly (e.g., SS2PL below).
• Precedence graph (or Serializability graph, Conflict
graph) cycle elimination
• Two-phase locking (2PL)
• Timestamp ordering (TO)
In a blocking method typically a context switching occurs
upon conflict, with (additional) incurred overhead. Oth• Serializable snapshot isolation[5] (SerializableSI)
erwise blocked transactions’ related computing resources
remain idle, unutilized, which may be a worse alternative.
The above (conflict) serializability techniques in their When conflicts do not occur frequently, optimistic methgeneral form do not provide recoverability. Special en- ods typically have an advantage. With different transhancements are needed for adding recoverability.
actions loads (mixes of transaction types) one technique
type (i.e., either optimistic or pessimistic) may provide
better performance than the other.
Optimistic versus pessimistic techniques ConcurUnless schedule classes are inherently blocking (i.e., they
rency control techniques are of three major types:
cannot be implemented without data-access operations
blocking; e.g., 2PL, SS2PL and SCO above; see chart),
1. Pessimistic: In Pessimistic concurrency control a
they can be implemented also using optimistic techniques
transaction blocks data access operations of other
(e.g., Serializability, Recoverability).
transactions upon conflicts, and conflicts are nonmaterialized until blocking is removed. This is done
to ensure that operations that may violate serializ- Serializable multi-version concurrency control
ability (and in practice also recoverability) do not
See also Multiversion concurrency control
(partial coverage)
2. Optimistic: In Optimistic concurrency control data
access operations of other transactions are not
Snapshot isolation
blocked upon conflicts, and conflicts are immediately materialized. When the transaction reaches
the ready state, i.e., its running state has been com- Multi-version concurrency control (MVCC) is a completed, possible serializability (and in practice also mon way today to increase concurrency and performance
recoverability) violation by the transaction’s oper- by generating a new version of a database object each
ations (relatively to other running transactions) is time the object is written, and allowing transactions’ read
checked: If violation has occurred, the transac- operations of several last relevant versions (of each obtion is typically aborted (sometimes aborting an- ject), depending on scheduling method. MVCC can
other transaction to handle serializability violation be combined with all the serializability techniques listed
above (except SerializableSI which is originally MVCC
is preferred). Otherwise it is committed.
based). It is utilized in most general-purpose DBMS
3. Semi-optimistic: Mechanisms that mix blocking in products.
certain situations with not blocking in other sitMVCC is especially popular nowadays through the reuations and employ both materialized and nonlaxed serializability (see above) method Snapshot isolation
materialized conflicts
(SI) which provides better performance than most known
serializability mechanisms (at the cost of possible serialThe main differences between the technique types is the izability violation in certain cases). SerializableSI, which
conflict types that are generated by them. A pessimistic is an efficient enhancement of SI to make it serializable,
method blocks a transaction operation upon conflict and is intended to provide an efficient serializable solution.
generates a non-materialized conflict, while an optimistic SerializableSI has been analyzed[5][6] via a general theory
method does not block and generates a materialized con- of MVCC
flict. A semi-optimistic method generates both conflict
types. Both conflict types are generated by the chronological orders in which transaction operations are invoked, 4.1.5 Distributed serializability
independently of the type of conflict. A cycle of committed transactions (with materialized conflicts) in the Overview
precedence graph (conflict graph) represents a serializability violation, and should be avoided for maintain- Distributed serializability is the serializability of a
ing serializability. A cycle of (non-materialized) con- schedule of a transactional distributed system (e.g., a
flicts in the wait-for graph represents a deadlock situation, distributed database system). Such system is characterwhich should be resolved by breaking the cycle. Both ized by distributed transactions (also called global transcycle types result from conflicts, and should be broken. actions), i.e., transactions that span computer processes
At any technique type conflicts should be detected and (a process abstraction in a general sense, depending on
considered, with similar overhead for both materialized computing environment; e.g., operating system's thread)
and non-materialized conflicts (typically by using mecha- and possibly network nodes. A distributed transaction
nisms like locking, while either blocking for locks, or not comprises more than one local sub-transactions that each
blocking but recording conflict for materialized conflicts). has states as described above for a database transaction.
A local sub-transaction comprises a single process, or 4.1.7 Notes
more processes that typically fail together (e.g., in a single
processor core). Distributed transactions imply a need in [1] Philip A. Bernstein, Vassos Hadzilacos, Nathan Goodman
(1987): Concurrency Control and Recovery in Database
Atomic commit protocol to reach consensus among its loSystems (free PDF download), Addison Wesley Publishing
cal sub-transactions on whether to commit or abort. Such
Company, ISBN 0-201-10715-5
protocols can vary from a simple (one-phase) hand-shake
among processes that fail together, to more sophisticated [2] Gerhard Weikum,
Gottfried Vossen (2001):
protocols, like Two-phase commit, to handle more comTransactional Information Systems, Elsevier, ISBN
plicated cases of failure (e.g., process, node, communi1-55860-508-8
cation, etc. failure). Distributed serializability is a major
goal of distributed concurrency control for correctness. [3] Maurice Herlihy and J. Eliot B. Moss. Transactional
memory: architectural support for lock-free data strucWith the proliferation of the Internet, Cloud computing,
tures. Proceedings of the 20th annual international symGrid computing, and small, portable, powerful computposium on Computer architecture (ISCA '93). Volume
ing devices (e.g., smartphones) the need for effective dis21, Issue 2, May 1993.
tributed serializability techniques to ensure correctness in
[4] Gray, J.; Helland, P.; O’Neil, P.; Shasha, D. (1996).
and among distributed applications seems to increase.
The dangers of replication and a solution (PDF). Pro-
Distributed serializability is achieved by implementceedings of the 1996 ACM SIGMOD International
ing distributed versions of the known centralized
Conference on Management of Data. pp. 173–182.
techniques.[1][2] Typically all such distributed versions redoi:10.1145/233269.233330.
quire utilizing conflict information (either of materialized or non-materialized conflicts, or equivalently, trans- [5] Michael J. Cahill, Uwe Röhm, Alan D. Fekete (2008):
“Serializable isolation for snapshot databases”, Proceedaction precedence or blocking information; conflict seings of the 2008 ACM SIGMOD international conferrializability is usually utilized) that is not generated loence on Management of data, pp. 729-738, Vancoucally, but rather in different processes, and remote lover, Canada, June 2008, ISBN 978-1-60558-102-6 (SIGcations. Thus information distribution is needed (e.g.,
MOD 2008 best paper award)
precedence relations, lock information, timestamps, or
tickets). When the distributed system is of a relatively [6] Alan Fekete (2009), “Snapshot Isolation and Serializable
Execution”, Presentation, Page 4, 2009, The university of
small scale, and message delays across the system are
Sydney (Australia). Retrieved 16 September 2009
small, the centralized concurrency control methods can
be used unchanged, while certain processes or nodes in
the system manage the related algorithms. However, in
4.1.8 References
a large-scale system (e.g., Grid and Cloud), due to the
distribution of such information, substantial performance
• Philip A. Bernstein, Vassos Hadzilacos, Nathan
penalty is typically incurred, even when distributed verGoodman (1987): Concurrency Control and Recovsions of the methods (Vs. centralized) are used, primarery in Database Systems, Addison Wesley Publishing
ily due to computer and communication latency. Also,
Company, ISBN 0-201-10715-5
when such information is distributed, related techniques
typically do not scale well. A well-known example with
• Gerhard Weikum, Gottfried Vossen (2001):
scalability problems is a distributed lock manager, which
Transactional Information Systems, Elsevier, ISBN
distributes lock (non-materialized conflict) information
across the distributed system to implement locking techniques.
4.2 Isolation (database systems)
See also
In database systems, isolation determines how transaction integrity is visible to other users and systems. For
• Strong strict two-phase locking (SS2PL or Rigor- example, when a user is creating a Purchase Order and
has created the header, but not the Purchase Order lines,
is the header available for other systems/users, carrying
• Making snapshot isolation serializable[5] in Snapshot out concurrent operations (such as a report on Purchase
Orders), to see?
A lower isolation level increases the ability of many users
• Global serializability, where the Global serializabil- to access data at the same time, but increases the numity problem and its proposed solutions are described. ber of concurrency effects (such as dirty reads or lost updates) users might encounter. Conversely, a higher iso• Linearizability, a more general concept in lation level reduces the types of concurrency effects that
concurrent computing
users may encounter, but requires more system resources
and increases the chances that one transaction will block logic for the application to function correctly.
Most DBMSs offer a number of transaction isolation levIsolation is typically defined at database level as a prop- els, which control the degree of locking that occurs when
erty that defines how/when the changes made by one op- selecting data. For many database applications, the maeration become visible to other. On older systems, it may jority of database transactions can be constructed to avoid
be implemented systemically, for example through the requiring high isolation levels (e.g. SERIALIZABLE
use of temporary tables. In two-tier systems, a Trans- level), thus reducing the locking overhead for the sysaction Processing (TP) manager is required to maintain tem. The programmer must carefully analyze database
isolation. In n-tier systems (such as multiple websites at- access code to ensure that any relaxation of isolation does
tempting to book the last seat on a flight), a combination not cause software bugs that are difficult to find. Conof stored procedures and transaction management is re- versely, if higher isolation levels are used, the possibility
quired to commit the booking and send confirmation to of deadlock is increased, which also requires careful analthe customer.[2]
ysis and programming techniques to avoid.
Isolation is one of the ACID (Atomicity, Consistency, The isolation levels defined by the ANSI/ISO SQL stanIsolation, Durability) properties.
dard are listed as follows.
Concurrency control
Concurrency control comprises the underlying mechanisms in a DBMS which handles isolation and guarantees related correctness. It is heavily utilized by the
database and storage engines (see above) both to guarantee the correct execution of concurrent transactions, and
(different mechanisms) the correctness of other DBMS
processes. The transaction-related mechanisms typically
constrain the database data access operations’ timing
(transaction schedules) to certain orders characterized as
the serializability and recoverability schedule properties.
Constraining database access operation execution typically means reduced performance (rates of execution),
and thus concurrency control mechanisms are typically
designed to provide the best performance possible under
the constraints. Often, when possible without harming
correctness, the serializability property is compromised
for better performance. However, recoverability cannot
be compromised, since such typically results in a quick
database integrity violation.
This is the highest isolation level.
With a lock-based concurrency control DBMS implementation, serializability requires read and write locks
(acquired on selected data) to be released at the end of the
transaction. Also range-locks must be acquired when a
SELECT query uses a ranged WHERE clause, especially
to avoid the phantom reads phenomenon (see below).
When using non-lock based concurrency control, no locks
are acquired; however, if the system detects a write collision among several concurrent transactions, only one of
them is allowed to commit. See snapshot isolation for
more details on this topic.
Repeatable reads
In this isolation level, a lock-based concurrency control
DBMS implementation keeps read and write locks (acquired on selected data) until the end of the transacTwo-phase locking is the most common transaction con- tion. However, range-locks are not managed, so phantom
currency control method in DBMSs, used to provide both reads can occur.
serializability and recoverability for correctness. In order to access a database object a transaction first needs
to acquire a lock for this object. Depending on the ac- Read committed
cess operation type (e.g., reading or writing an object)
and on the lock type, acquiring the lock may be blocked In this isolation level, a lock-based concurrency control
and postponed, if another transaction is holding a lock for DBMS implementation keeps write locks (acquired on
selected data) until the end of the transaction, but read
that object.
locks are released as soon as the SELECT operation is
performed (so the non-repeatable reads phenomenon can
occur in this isolation level, as discussed below). As in
4.2.2 Isolation levels
the previous level, range-locks are not managed.
Of the four ACID properties in a DBMS (Database Management System), the isolation property is the one most
often relaxed. When attempting to maintain the highest
level of isolation, a DBMS usually acquires locks on data
or implements multiversion concurrency control, which
may result in a loss of concurrency. This requires adding
Putting it in simpler words, read committed is an isolation level that guarantees that any data read is committed
at the moment it is read. It simply restricts the reader
from seeing any intermediate, uncommitted, 'dirty' read.
It makes no promise whatsoever that if the transaction reissues the read, it will find the same data; data is free to
change after it is read.
Read uncommitted
In the following examples, two transactions take place. In
the first, Query 1 is performed. Then, in the second transaction, Query 2 is performed and committed. Finally, in
the first transaction, Query 1 is performed again.
This is the lowest isolation level. In this level, dirty reads The queries use the following data table:
are allowed, so one transaction may see not-yet-committed
changes made by other transactions.
Dirty reads
Since each isolation level is stronger than those below,
in that no higher isolation level allows an action forbid- A dirty read (aka uncommitted dependency) occurs when
den by a lower one, the standard permits a DBMS to run a transaction is allowed to read data from a row that has
a transaction at an isolation level stronger than that re- been modified by another running transaction and not yet
quested (e.g., a “Read committed” transaction may actu- committed.
ally be performed at a “Repeatable read” isolation level).
Dirty reads work similarly to non-repeatable reads; however, the second transaction would not need to be committed for the first query to return a different result. The
4.2.3 Default isolation level
only thing that may be prevented in the READ UNCOMThe default isolation level of different DBMS's varies MITTED isolation level is updates appearing out of order
quite widely. Most databases that feature transactions al- in the results; that is, earlier updates will always appear
low the user to set any isolation level. Some DBMS’s also in a result set before later updates.
require additional syntax when performing a SELECT In our example, Transaction 2 changes a row, but does
statement to acquire locks (e.g. SELECT ... FOR UP- not commit the changes. Transaction 1 then reads the
DATE to acquire exclusive write locks on accessed rows). uncommitted data. Now if Transaction 2 rolls back its
However, the definitions above have been criticized [3] changes (already read by Transaction 1) or updates difas being ambiguous, and as not accurately reflecting the ferent changes to the database, then the view of the data
may be wrong in the records of Transaction 1.
isolation provided by many databases:
This paper shows a number of weaknesses in
the anomaly approach to defining isolation levels. The three ANSI phenomena are ambiguous. Even their broadest interpretations do
not exclude anomalous behavior. This leads
to some counter-intuitive results. In particular,
lock-based isolation levels have different characteristics than their ANSI equivalents. This
is disconcerting because commercial database
systems typically use locking. Additionally,
the ANSI phenomena do not distinguish among
several isolation levels popular in commercial
But in this case no row exists that has an id of 1 and an
age of 21.
Non-repeatable reads
A non-repeatable read occurs, when during the course
of a transaction, a row is retrieved twice and the values
within the row differ between reads.
Non-repeatable reads phenomenon may occur in a lockbased concurrency control method when read locks are
not acquired when performing a SELECT, or when the
acquired locks on affected rows are released as soon as the
SELECT operation is performed. Under the multiversion
concurrency control method, non-repeatable reads may
There are also other criticisms concerning ANSI SQL’s occur when the requirement that a transaction affected by
isolation definition, in that it encourages implementors to a commit conflict must roll back is relaxed.
do “bad things":
In this example, Transaction 2 commits successfully,
which means that its changes to the row with id 1 should
... it relies in subtle ways on an assumption
become visible. However, Transaction 1 has already seen
that a locking schema is used for concurrency
a different value for age in that row. At the SERIALIZcontrol, as opposed to an optimistic or multiABLE and REPEATABLE READ isolation levels, the
version concurrency scheme. This implies that
DBMS must return the old value for the second SEthe proposed semantics are ill-defined.[4]
LECT. At READ COMMITTED and READ UNCOMMITTED, the DBMS may return the updated value; this
is a non-repeatable read.
Read phenomena
There are two basic strategies used to prevent nonThe ANSI/ISO standard SQL 92 refers to three differ- repeatable reads. The first is to delay the execution of
ent read phenomena when Transaction 1 reads data that Transaction 2 until Transaction 1 has committed or rolled
Transaction 2 might have changed.
back. This method is used when locking is used, and pro-
duces the serial schedule T1, T2. A serial schedule ex- locked, thus Query 2 would block until the first transachibits repeatable reads behaviour.
tion was committed. In REPEATABLE READ mode,
In the other strategy, as used in multiversion concurrency the range would not be locked, allowing the record to be
control, Transaction 2 is permitted to commit first, which inserted and the second execution of Query 1 to include
provides for better concurrency. However, Transaction 1, the new row in its results.
which commenced prior to Transaction 2, must continue
to operate on a past version of the database — a snap4.2.5 Isolation Levels, Read Phenomena
shot of the moment it was started. When Transaction 1
and Locks
eventually tries to commit, the DBMS checks if the result
of committing Transaction 1 would be equivalent to the
schedule T1, T2. If it is, then Transaction 1 can proceed. Isolation Levels vs Read Phenomena
If it cannot be seen to be equivalent, however, TransacAnomaly Serializable is not the same as Serializable.
tion 1 must roll back with a serialization failure.
That is, it is necessary, but not sufficient that a SerializUsing a lock-based concurrency control method, at the able schedule should be free of all three phenomena types.
REPEATABLE READ isolation mode, the row with ID See [1] below.
= 1 would be locked, thus blocking Query 2 until the
first transaction was committed or rolled back. In READ “may occur” means that the isolation level suffers that
COMMITTED mode, the second time Query 1 was exe- phenomenon, while "-" means that it does not suffer it.
cuted, the age would have changed.
Under multiversion concurrency control, at the SERIALIZABLE isolation level, both SELECT queries see a
snapshot of the database taken at the start of Transaction 1. Therefore, they return the same data. However,
if Transaction 1 then attempted to UPDATE that row as
well, a serialization failure would occur and Transaction
1 would be forced to roll back.
Isolation Levels vs Lock Duration
In lock-based concurrency control, isolation level determines the duration that locks are held.
“C” - Denotes that locks are held until the transaction
“S” - Denotes that the locks are held only during the currently executing statement. Note that if locks are released
At the READ COMMITTED isolation level, each query
after a statement, the underlying data could be changed by
sees a snapshot of the database taken at the start of each
another transaction before the current transaction comquery. Therefore, they each see different data for the
mits, thus creating a violation.
updated row. No serialization failure is possible in this
mode (because no promise of serializability is made), and
Transaction 1 will not have to be retried.
4.2.6 See also
Phantom reads
A phantom read occurs when, in the course of a transaction, two identical queries are executed, and the collection
of rows returned by the second query is different from the
This can occur when range locks are not acquired on performing a SELECT ... WHERE operation. The phantom
reads anomaly is a special case of Non-repeatable reads
when Transaction 1 repeats a ranged SELECT ... WHERE
query and, between both operations, Transaction 2 creates (i.e. INSERT) new rows (in the target table) which
fulfill that WHERE clause.
Note that Transaction 1 executed the same query twice.
If the highest level of isolation were maintained, the same
set of rows should be returned both times, and indeed that
is what is mandated to occur in a database operating at
the SQL SERIALIZABLE isolation level. However, at
the lesser isolation levels, a different set of rows may be
returned the second time.
In the SERIALIZABLE isolation mode, Query 1 would
result in all records with age in the range 10 to 30 being
• Atomicity
• Consistency
• Durability
• Lock (database)
• Optimistic concurrency control
• Relational Database Management System
• Snapshot isolation
4.2.7 References
[1] “Isolation Levels in the Database Engine”, Technet,
[2] “The Architecture of Transaction Processing Systems”,
Chapter 23, Evolution of Processing Systems, Department of Computer Science, Stony Brook University,
retrieved 20 March 2014,
[3] “A Critique of ANSI SQL Isolation Levels” (PDF). Retrieved 29 July 2012.
[4] salesforce (2010-12-06). “Customer testimonials (SimpleGeo, CLOUDSTOCK 2010)".
DataStax. Retrieved 2010-03-09. (see above at about
13:30 minutes of the webcast!)
External links
4.3.1 Purpose
Databases and other data stores which treat the integrity
of data as paramount often include the ability to handle transactions to maintain the integrity of data. A single transaction consists of one or more independent units
of work, each reading and/or writing information to a
database or other data store. When this happens it is often important to ensure that all such processing leaves the
database or data store in a consistent state.
• Oracle® Database Concepts, chapter 13 Data ConExamples from double-entry accounting systems often ilcurrency and Consistency, Preventable Phenomena
lustrate the concept of transactions. In double-entry acand Transaction Isolation Levels
counting every debit requires the recording of an asso• Oracle® Database SQL Reference, chapter 19 ciated credit. If one writes a check for $100 to buy
SQL Statements: SAVEPOINT to UPDATE, SET groceries, a transactional double-entry accounting system
must record the following two entries to cover the single
• in
1. Debit $100 to Groceries Expense Account
2. Credit $100 to Checking Account
• in Spring Framework: @Transactional, Isolation
• P.Bailis. When is “ACID” ACID? Rarely
A transactional system would make both entries pass or
both entries would fail. By treating the recording of multiple entries as an atomic transactional unit of work the
system maintains the integrity of the data recorded. In
4.3 Database transaction
other words, nobody ends up with a situation in which a
debit is recorded but no associated credit is recorded, or
A transaction symbolizes a unit of work performed
vice versa.
within a database management system (or similar system) against a database, and treated in a coherent and reliable way independent of other transactions. A transaction 4.3.2 Transactional databases
generally represents any change in database. Transactions
in a database environment have two main purposes:
A transactional database is a DBMS where write trans1. To provide reliable units of work that allow correct
recovery from failures and keep a database consistent even in cases of system failure, when execution
stops (completely or partially) and many operations
upon a database remain uncompleted, with unclear
actions on the database are able to be rolled back if they
are not completed properly (e.g. due to power or connectivity loss).
Most modern relational database management systems
fall into the category of databases that support transactions.
In a database system a transaction might consist of one
or more data-manipulation statements and queries, each
reading and/or writing information in the database. Users
of database systems consider consistency and integrity of
data as highly important. A simple transaction is usually issued to the database system in a language like SQL
A database transaction, by definition, must be atomic, wrapped in a transaction, using a pattern similar to the
consistent, isolated and durable.[1] Database practition- following:
ers often refer to these properties of database transactions
using the acronym ACID.
1. Begin the transaction
2. To provide isolation between programs accessing a
database concurrently. If this isolation is not provided, the programs’ outcomes are possibly erroneous.
Transactions provide an “all-or-nothing” proposition,
stating that each work-unit performed in a database must
either complete in its entirety or have no effect whatsoever. Further, the system must isolate each transaction
from other transactions, results must conform to existing
constraints in the database, and transactions that complete
successfully must get written to durable storage.
2. Execute a set of data manipulations and/or queries
3. If no errors occur then commit the transaction and
end it
4. If errors occur then rollback the transaction and end
If no errors occurred during the execution of the transaction then the system commits the transaction. A transaction commit operation applies all data manipulations
within the scope of the transaction and persists the results
to the database. If an error occurs during the transaction,
or if the user specifies a rollback operation, the data manipulations within the transaction are not persisted to the
database. In no case can a partial transaction be committed to the database since that would leave the database in
an inconsistent state.
After starting a transaction, database records or objects
are locked, either read-only or read-write. Actual reads
and writes can then occur. Once the user (and application) is happy, any changes are committed or rolled-back
atomically, such that at the end of the transaction there is
no inconsistency.
4.3.4 Distributed transactions
Database systems implement distributed transactions as
Internally, multi-user databases store and process transtransactions against multiple applications or hosts. A disactions, often by using a transaction ID or XID.
tributed transaction enforces the ACID properties over
There are multiple varying ways for transactions to be im- multiple systems or data stores, and might include sysplemented other than the simple way documented above. tems such as databases, file systems, messaging systems,
Nested transactions, for example, are transactions which and other applications. In a distributed transaction a cocontain statements within them that start new transactions ordinating service ensures that all parts of the transaction
(i.e. sub-transactions). Multi-level transactions are a vari- are applied to all relevant systems. As with database and
ant of nested transactions where the sub-transactions take other transactions, if any part of the transaction fails, the
place at different levels of a layered system architecture entire transaction is rolled back across all affected sys(e.g., with one operation at the database-engine level, one tems.
operation at the operating-system level) [2] Another type
of transaction is the compensating transaction.
4.3.5 Transactional filesystems
The Namesys Reiser4 filesystem for Linux[3] supports
transactions, and as of Microsoft Windows Vista, the MiTransactions are available in most SQL database im- crosoft NTFS filesystem[4] supports distributed transacplementations, though with varying levels of robustness. tions across networks.
(MySQL, for example, does not support transactions in
the MyISAM storage engine, which was its default stor4.3.6 See also
age engine before version 5.5.)
A transaction is typically started using the com• Concurrency control
mand BEGIN (although the SQL standard specifies
START TRANSACTION). When the system processes
a COMMIT statement, the transaction ends with success- 4.3.7 References
ful completion. A ROLLBACK statement can also end
the transaction, undoing any work performed since BE- [1] A transaction is a group of operations that are atomic, conGIN TRANSACTION. If autocommit was disabled ussistent, isolated, and durable (ACID).
ing START TRANSACTION, autocommit will also be
[2] Beeri, C., Bernstein, P.A., and Goodman, N. A model for
re-enabled at the transaction’s end.
concurrency in nested transactions systems. Journal of the
One can set the isolation level for individual transactional
ACM, 36(1):230-269, 1989
operations as well as globally. At the READ COMMITTED level, the result of any work done after a transac- [3]
tion has commenced, but before it has ended, will remain
[4] “MSDN Library”. Retrieved 16 October 2014.
invisible to other database-users until it has ended. At
the lowest level (READ UNCOMMITTED), which may
occasionally be used to ensure high concurrency, such 4.3.8 Further reading
changes will be visible.
Object databases
Relational databases traditionally comprise tables with
fixed size fields and thus records. Object databases comprise variable sized blobs (possibly incorporating a mimetype or serialized). The fundamental similarity though is
the start and the commit or rollback.
• Philip A. Bernstein, Eric Newcomer (2009):
Principles of Transaction Processing, 2nd Edition,
Morgan Kaufmann (Elsevier), ISBN 978-1-55860623-4
• Gerhard Weikum, Gottfried Vossen (2001), Transactional information systems: theory, algorithms,
and the practice of concurrency control and recovery, Morgan Kaufmann, ISBN 1-55860-508-8
External links
• c2:TransactionProcessing
4.4 Transaction processing
For other uses, see Transaction (disambiguation).
This article is about the principles of transaction processing. For specific implementations, see Transaction
processing system.
In computer science, transaction processing is information processing that is divided into individual, indivisible operations called transactions. Each transaction must
succeed or fail as a complete unit; it can never be only
partially complete.
For example, when you purchase a book from an online
bookstore, you exchange money (in the form of credit)
for a book. If your credit is good, a series of related
operations ensures that you get the book and the bookstore gets your money. However, if a single operation in
the series fails during the exchange, the entire exchange
fails. You do not get the book and the bookstore does not
get your money. The technology responsible for making
the exchange balanced and predictable is called transaction processing. Transactions ensure that data-oriented
resources are not permanently updated unless all operations within the transactional unit complete successfully.
By combining a set of related operations into a unit that
either completely succeeds or completely fails, one can
simplify error recovery and make one’s application more
checking account by $700. If one operation succeeds but
the other does not, the books of the bank will not balance at the end of the day. There must therefore be a
way to ensure that either both operations succeed or both
fail, so that there is never any inconsistency in the bank’s
database as a whole.
Transaction processing links multiple individual operations in a single, indivisible transaction, and ensures that
either all operations in a transaction are completed without error, or none of them are. If some of the operations
are completed but errors occur when the others are attempted, the transaction-processing system “rolls back”
all of the operations of the transaction (including the successful ones), thereby erasing all traces of the transaction
and restoring the system to the consistent, known state
that it was in before processing of the transaction began.
If all operations of a transaction are completed successfully, the transaction is committed by the system, and all
changes to the database are made permanent; the transaction cannot be rolled back once this is done.
Transaction processing guards against hardware and software errors that might leave a transaction partially completed. If the computer system crashes in the middle of
a transaction, the transaction processing system guarantees that all operations in any uncommitted transactions
are cancelled.
Since most, though not necessarily all, transaction processing today is interactive the term is often treated as
synonymous with online transaction processing.
Generally, transactions are issued concurrently. If they
overlap (i.e. need to touch the same portion of the
database), this can create conflicts. For example, if the
customer mentioned in the example above has $150 in
his savings account and attempts to transfer $100 to a different person while at the same time moving $100 to the
checking account, only one of them can succeed. However, forcing transactions to be processed sequentially
is inefficient. Therefore, concurrent implementations of
transaction processing is programmed to guarantee that
the end result reflects a conflict-free outcome, the same
as could be reached if executing the transactions sequentially in any order (a property called serializability). In
our example, this means that no matter which transaction
was issued first, either the transfer to a different person
or the move to the checking account succeeds, while the
other one fails.
4.4.2 Methodology
Transaction processing systems consist of computer hardware and software hosting a transaction-oriented application that performs the routine transactions necessary to
conduct business. Examples include systems that manage
sales order entry, airline reservations, payroll, employee
records, manufacturing, and shipping.
Transaction processing is designed to maintain a system’s Integrity (typically a database or some modern
filesystems) in a known, consistent state, by ensuring that
interdependent operations on the system are either all
completed successfully or all canceled successfully.
The basic principles of all transaction-processing systems
are the same. However, the terminology may vary from
one transaction-processing system to another, and the
terms used below are not necessarily universal.
For example, consider a typical banking transaction that
involves moving $700 from a customer’s savings account Rollback
to a customer’s checking account. This transaction involves at least two separate operations in computer terms: Main article: Rollback (data management)
debiting the savings account by $700, and crediting the
Transaction-processing systems ensure database integrity
by recording intermediate states of the database as it is
modified, then using these records to restore the database
to a known state if a transaction cannot be committed.
For example, copies of information on the database prior
to its modification by a transaction are set aside by the
system before the transaction can make any modifications
(this is sometimes called a before image). If any part of
the transaction fails before it is committed, these copies
are used to restore the database to the state it was in before
the transaction began.
they can detect.
Main article: ACID
It is also possible to keep a separate journal of all modifications to a database management system. (sometimes
called after images). This is not required for rollback
of failed transactions but it is useful for updating the
database management system in the event of a database
failure, so some transaction-processing systems provide
it. If the database management system fails entirely, it
must be restored from the most recent back-up. The
back-up will not reflect transactions committed since the
back-up was made. However, once the database management system is restored, the journal of after images
can be applied to the database (rollforward) to bring the
database management system up to date. Any transactions in progress at the time of the failure can then be
rolled back. The result is a database in a consistent,
known state that includes the results of all transactions
committed up to the moment of failure.
Main article: Deadlock
In some cases, two transactions may, in the course of
their processing, attempt to access the same portion of
a database at the same time, in a way that prevents them
from proceeding. For example, transaction A may access
portion X of the database, and transaction B may access
portion Y of the database. If, at that point, transaction
A then tries to access portion Y of the database while
transaction B tries to access portion X, a deadlock occurs,
and neither transaction can move forward. Transactionprocessing systems are designed to detect these deadlocks
when they occur. Typically both transactions will be cancelled and rolled back, and then they will be started again
in a different order, automatically, so that the deadlock
doesn't occur again. Or sometimes, just one of the deadlocked transactions will be cancelled, rolled back, and automatically restarted after a short delay.
Deadlocks can also occur among three or more transactions. The more transactions involved, the more difficult
they are to detect, to the point that transaction processing systems find there is a practical limit to the deadlocks
Compensating transaction
In systems where commit and rollback mechanisms are
not available or undesirable, a compensating transaction
is often used to undo failed transactions and restore the
system to a previous state.
4.4.3 ACID criteria
Jim Gray defined properties of a reliable transaction system in the late 1970s under the acronym ACID — atomicity, consistency, isolation, and durability.[1]
Main article: Atomicity (database systems)
A transaction’s changes to the state are atomic: either all
happen or none happen. These changes include database
changes, messages, and actions on transducers.
Consistency: A transaction is a correct transformation of
the state. The actions taken as a group do not violate any
of the integrity constraints associated with the state.
Even though transactions execute concurrently, it appears
to each transaction T, that others executed either before
T or after T, but not both.
Once a transaction completes successfully (commits), its
changes to the state survive failures.
4.4.4 Benefits
Transaction processing has these benefits:
• It allows sharing of computer resources among many
• It shifts the time of job processing to when the computing resources are less busy
• It avoids idling the computing resources without 4.4.8 Further reading
minute-by-minute human interaction and supervi• Gerhard Weikum, Gottfried Vossen, Transactional
information systems: theory, algorithms, and the
• It is used on expensive classes of computers to help
practice of concurrency control and recovery, Moramortize the cost by keeping high rates of utilization
gan Kaufmann, 2002, ISBN 1-55860-508-8
of those expensive resources
• Jim Gray, Andreas Reuter, Transaction Processing
— Concepts and Techniques, 1993, Morgan Kaufmann, ISBN 1-55860-190-2
4.4.5 Implementations
Main article: Transaction processing system
Standard transaction-processing software, notably IBM's
Information Management System, was first developed in
the 1960s, and was often closely coupled to particular
database management systems. Client–server computing
implemented similar principles in the 1980s with mixed
success. However, in more recent years, the distributed
client–server model has become considerably more difficult to maintain. As the number of transactions grew in
response to various online services (especially the Web),
a single distributed database was not a practical solution.
In addition, most online systems consist of a whole suite
of programs operating together, as opposed to a strict
client–server model where the single server could handle the transaction processing. Today a number of transaction processing systems are available that work at the
inter-program level and which scale to large systems, including mainframes.
One well-known (and open) industry standard is the
X/Open Distributed Transaction Processing (DTP) (see
also JTA the Java Transaction API). However, proprietary transaction-processing environments such as IBM’s
CICS are still very popular, although CICS has evolved
to include open industry standards as well.
The term 'Extreme Transaction Processing' (XTP) has
been used to describe transaction processing systems
with uncommonly challenging requirements, particularly
throughput requirements (transactions per second). Such
systems may be implemented via distributed or cluster
style architectures.
[1] Gray, Jim; Reuter, Andreas. “Transaction Processing Concepts and Techniques (Powerpoint)". Retrieved Nov
12, 2012.
External links
• Nuts and Bolts of Transaction Processing (1999)
• Managing Transaction Processing for SQL Database
• Transaction Processing
• Philip A. Bernstein, Eric Newcomer, Principles of
Transaction Processing, 1997, Morgan Kaufmann,
ISBN 1-55860-415-4
• Ahmed K. Elmagarmid (Editor), Transaction Models for Advanced Database Applications, MorganKaufmann, 1992, ISBN 1-55860-214-3
Chapter 5
5.1 Journaling file system
example by a tool such as fsck (the file system checker).[2]
This must typically be done before the file system is next
For the IBM Journaled File System, see JFS (file system). mounted for read-write access. If the file system is large
and if there is relatively little I/O bandwidth, this can take
a long time and result in longer downtimes if it blocks the
A journaling file system is a file system that keeps track rest of the system from coming back online.
of changes not yet committed to the file system’s main
part by recording the intentions of such changes in a data To prevent this, a journaled file system allocates a spestructure known as a "journal", which is usually a circular cial area—the journal—in which it records the changes it
log. In the event of a system crash or power failure, such will make ahead of time. After a crash, recovery simply
file systems can be brought back online quicker with lower involves reading the journal from the file system and replaying changes from this journal until the file system is
likelihood of becoming corrupted.[1][2]
consistent again. The changes are thus said to be atomic
Depending on the actual implementation, a journaling file (not divisible) in that they either succeed (succeeded origsystem may only keep track of stored metadata, resulting inally or are replayed completely during recovery), or are
in improved performance at the expense of increased pos- not replayed at all (are skipped because they had not yet
sibility for data corruption. Alternatively, a journaling file been completely written to the journal before the crash
system may track both stored data and related metadata, occurred).
while some implementations allow selectable behavior in
this regard.[3]
5.1.2 Techniques
Updating file systems to reflect changes to files and directories usually requires many separate write operations.
This makes it possible for an interruption (like a power
failure or system crash) between writes to leave data
structures in an invalid intermediate state.[1]
For example, deleting a file on a Unix file system involves
three steps:[4]
Some file systems allow the journal to grow, shrink and
be re-allocated just as a regular file, while others put the
journal in a contiguous area or a hidden file that is guaranteed not to move or change size while the file system is
mounted. Some file systems may also allow external journals on a separate device, such as a solid-state drive or
battery-backed non-volatile RAM. Changes to the journal may themselves be journaled for additional redundancy, or the journal may be distributed across multiple
physical volumes to protect against device failure.
1. Removing its directory entry.
The internal format of the journal must guard against
crashes while the journal itself is being written to. Many
2. Release the inode to the pool of free inodes.
journal implementations (such as the JBD2 layer in ext4)
3. Return all used disk blocks to the pool of free disk bracket every change logged with a checksum, on the understanding that a crash would leave a partially written
change with a missing (or mismatched) checksum that
If a crash occurs after step 1 and before step 2, there can simply be ignored when replaying the journal at next
will be an orphaned inode and hence a storage leak. On remount.
the other hand, if only step 2 is performed first before
the crash, the not-yet-deleted file will be marked free and
Physical journals
possibly be overwritten by something else.
Detecting and recovering from such inconsistencies nor- A physical journal logs an advance copy of every block
mally requires a complete walk of its data structures, for that will later be written to the main file system. If there is
a crash when the main file system is being written to, the cache at certain points in the journal (called barriers in
write can simply be replayed to completion when the file ext3 and ext4).[7]
system is next mounted. If there is a crash when the write
is being logged to the journal, the partial write will have
a missing or mismatched checksum and can be ignored at 5.1.3 Alternatives
next mount.
Soft updates
Physical journals impose a significant performance
penalty because every changed block must be commit- Some UFS implementations avoid journaling and instead
ted twice to storage, but may be acceptable when absolute implement soft updates: they order their writes in such
fault protection is required.[5]
a way that the on-disk file system is never inconsistent,
or that the only inconsistency that can be created in the
event of a crash is a storage leak. To recover from these
Logical journals
leaks, the free space map is reconciled against a full walk
A logical journal stores only changes to file metadata in of the file system at next mount. This garbage collection
the journal, and trades fault tolerance for substantially is usually done in the background.
better write performance.[6] A file system with a logical
journal still recovers quickly after a crash, but may allow
Log-structured file systems
unjournaled file data and journaled metadata to fall out
of sync with each other, causing data corruption.
In log-structured file systems, the write-twice penalty
For example, appending to a file may involve three sepa- does not apply because the journal itself is the file system:
it occupies the entire storage device and is structured so
rate writes to:
that it can be traversed as would a normal file system.
1. The file’s inode, to note in the file’s metadata that its
size has increased.
Copy-on-write file systems
2. The free space map, to mark out an allocation of
Full copy-on-write file systems (such as ZFS and Btrfs)
space for the to-be-appended data.
avoid in-place changes to file data by writing out the data
3. The newly allocated space, to actually write the ap- in newly allocated blocks, followed by updated metadata
that would point to the new data and disown the old, folpended data.
lowed by metadata pointing to that, and so on up to the
In a metadata-only journal, step 3 would not be logged. If superblock, or the root of the file system hierarchy. This
step 3 was not done, but steps 1 and 2 are replayed during has the same correctness-preserving properties as a journal, without the write-twice overhead.
recovery, the file will be appended with garbage.
Write hazards
The write cache in most operating systems sorts its writes
(using the elevator algorithm or some similar scheme)
to maximize throughput. To avoid an out-of-order write
hazard with a metadata-only journal, writes for file data
must be sorted so that they are committed to storage before their associated metadata. This can be tricky to implement because it requires coordination within the operating system kernel between the file system driver and
write cache. An out-of-order write hazard can also exist
if the underlying storage cannot write blocks atomically,
or does not honor requests to flush its write cache.
5.1.4 See also
• Comparison of file systems
• Database
• Intent log
• Journaled File System (JFS) – a file system made by
• Transaction processing
To complicate matters, many mass storage devices have
their own write caches, in which they may aggressively re- 5.1.5 References
order writes for better performance. (This is particularly
common on magnetic hard drives, which have large seek [1] Jones, M Tim (2008-06-04), Anatomy of Linux journaling
file systems, IBM DeveloperWorks, retrieved 2009-04-13
latencies that can be minimized with elevator sorting.)
Some journaling file systems conservatively assume such [2] Arpaci-Dusseau, Remzi H.; Arpaci-Dusseau, Andrea C.
write-reordering always takes place, and sacrifice perfor(2014-01-21), Crash Consistency: FSCK and Journaling
(PDF), Arpaci-Dusseau Books
mance for correctness by forcing the device to flush its
[3] “tune2fs(8) – Linux man page”. Retrieved
February 20, 2015.
[4] File Systems from Tanenbaum, A.S. (2008). Modern operating systems (3rd ed., pp. 287). Upper Saddle River,
NJ: Prentice Hall.
[5] Tweedie, Stephen (2000), “Ext3, journaling filesystem”,
Proceedings of the Ottawa Linux Symposium: 24–29
[6] Prabhakaran, Vijayan; Arpaci-Dusseau, Andrea C;
Arpaci-Dusseau, Remzi H, “Analysis and Evolution of
Journaling File Systems” (PDF), 2005 USENIX Annual
Technical Conference (USENIX Association).
[7] Corbet, Jonathan (2008-05-21), Barriers and journaling
filesystems, retrieved 2010-03-06
[8] Seltzer, Margo I; Ganger, Gregory R; McKusick, M Kirk,
“Journaling Versus Soft Updates: Asynchronous Metadata Protection in File Systems”, 2000 USENIX Annual
Technical Conference (USENIX Association).
5.2 Atomicity (database systems)
For other uses, see Atomicity (disambiguation).
In database systems, atomicity (or atomicness; from
Greek a-tomos, undividable) is one of the ACID
transaction properties. In an atomic transaction, a series of database operations either all occur, or nothing
occurs. The series of operations cannot be divided apart
and executed partially from each other, which makes the
series of operations “indivisible”, hence the name. A
guarantee of atomicity prevents updates to the database
occurring only partially, which can cause greater problems than rejecting the whole series outright. In other
words, atomicity means indivisibility and irreducibility.[1]
As a consequence, the transaction cannot be observed to
be in progress by another database client. At one moment in time, it has not yet happened, and at the next it
has already occurred in whole (or nothing happened if the
transaction was cancelled in progress).
The booking system does not consider it acceptable for a
customer to pay for a ticket without securing the seat, nor
to reserve the seat without payment succeeding.
Another example is that if one wants to transfer some
amount of money from one account to another, then the
user would start a procedure to do it. However, if a failure
occurs, then due to atomicity, the amount will either be
transferred completely or will not be even initiated. Thus
atomicity protects the user from losing money due to a
failed transaction.
5.2.2 Orthogonality
Atomicity does not behave completely orthogonally with
regard to the other ACID properties of the transactions. For example, isolation relies on atomicity to roll
back changes in the event of isolation failures such as
deadlock; consistency also relies on rollback in the event
of a consistency-violation by an illegal transaction. Finally, atomicity itself relies on durability to ensure the
atomicity of transactions even in the face of external failures.
As a result of this, failure to detect errors and roll back
the enclosing transaction may cause failures of isolation
and consistency.
5.2.3 Implementation
Typically, systems implement Atomicity by providing
some mechanism to indicate which transactions have
started and which finished; or by keeping a copy of the
data before any changes occurred (read-copy-update).
Several filesystems have developed methods for avoiding
the need to keep multiple copies of data, using journaling (see journaling file system). Databases usually implement this using some form of logging/journaling to
track changes. The system synchronizes the logs (often
the metadata) as necessary once the actual changes have
successfully taken place. Afterwards, crash recovery simply ignores incomplete entries. Although implementations vary depending on factors such as concurrency isThe etymology of the phrase originates in the Classical sues, the principle of atomicity — i.e. complete success
Greek concept of a fundamental and indivisible compo- or complete failure — remain.
nent; see atom.
Ultimately, any application-level implementation relies
on operating-system functionality. At the file-system
level, POSIX-compliant systems provide system calls
5.2.1 Examples
such as open(2) and flock(2) that allow applications to
atomically open or lock a file. At the process level,
An example of atomicity is ordering an airline ticket POSIX Threads provide adequate synchronization primwhere two actions are required: payment, and a seat itives.
reservation. The potential passenger must either:
The hardware level requires atomic operations such
1. both pay for and reserve a seat; OR
2. neither pay for nor reserve a seat.
as Test-and-set, Fetch-and-add, Compare-and-swap, or
Load-Link/Store-Conditional, together with memory
barriers. Portable operating systems cannot simply block
interrupts to implement synchronization, since hardware
that lacks actual concurrent execution such as hyperthreading or multi-processing is now extremely rare.
In NoSQL data stores with eventual consistency, the
atomicity is also weaker specified than in relational
database systems, and exists only in rows (i.e. column
See also
• Atomic operation
• Transaction processing
• Long-running transaction
• Read-copy-update
[1] “atomic operation”. Webopedia. Retrieved 2011-03-23. An operation during which
a processor can simultaneously read a location and write
it in the same bus operation. This prevents any other processor or I/O device from writing or reading memory until
the operation is complete.
[2] Olivier Mallassi (2010-06-09). “Let’s play with Cassandra… (Part 1/3)". OCTO
Talks!. Retrieved 2011-03-23. Atomicity is also weaker
than what we are used to in the relational world. Cassandra guarantees atomicity within a ColumnFamily so for all
the columns of a row.
Chapter 6
6.1 Lock (database)
6.2 Record locking
A lock, as a read lock or write lock, is used when
multiple users need to access a database concurrently.
This prevents data from being corrupted or invalidated
when multiple users try to read while others write to the
database. Any single user can only modify those database
records (that is, items in the database) to which they have
applied a lock that gives them exclusive access to the
record until the lock is released. Locking not only provides exclusivity to writes but also prevents (or controls)
reading of unfinished modifications (AKA uncommitted
Record locking is the technique of preventing simultaneous access to data in a database, to prevent inconsistent
The classic example is demonstrated by two bank clerks
attempting to update the same bank account for two different transactions. Clerks 1 and 2 both retrieve (i.e.,
copy) the account’s record. Clerk 1 applies and saves a
transaction. Clerk 2 applies a different transaction to his
saved copy, and saves the result, based on the original
record and his changes, overwriting the transaction entered by clerk 1. The record no longer reflects the first
A read lock can be used to prevent other users from read- transaction, as if it had never taken place.
ing a record (or page) which is being updated, so that oth- A simple way to prevent this is to lock the file whenever a
ers will not act upon soon-to-be-outdated information.
record is being modified by any user, so that no other user
can save data. This prevents records from being overwritten incorrectly, but allows only one record to be processed
at a time, locking out other users who need to edit records
6.1.1 Mechanisms for locking
at the same time.
There are two mechanisms for locking data in a database:
pessimistic locking, and optimistic locking. In pessimistic
locking a record or page is locked immediately when the
lock is requested, while in an optimistic lock the record
or page is only locked when the changes made to that
record are updated. The latter situation is only appropriate when there is less chance of someone needing to
access the record while it is locked; otherwise it cannot be
certain that the update will succeed because the attempt
to update the record will fail if another user updates the
record first. With pessimistic locking it is guaranteed that
the record will be updated.
To allow several users to edit a database table at the same
time and also prevent inconsistencies created by unrestricted access, a single record can be locked when retrieved for editing or updating. Anyone attempting to retrieve the same record for editing is denied write access
because of the lock (although, depending on the implementation, they may be able to view the record without
editing it). Once the record is saved or edits are canceled,
the lock is released. Records can never be saved so as to
overwrite other changes, preserving data integrity.
In database management theory, locking is used to implement isolation among multiple database users. This is
the “I” in the acronym ACID.
The degree of locking can be controlled by isolation level.
Change of a lock is called lock conversion and the lock A thorough and authoritative description of locking was
may be upgraded (lock upgrade) or downgraded (lock written by Jim Gray.
6.2.1 Granularity of locks
See also
• Race condition
If the bank clerks (to follow the illustration above) are
serving two customers, but their accounts are contained in
one ledger, then the entire ledger, or one or more database
tables, would need to be made available for editing to the
clerks in order for each to complete a transaction, one at list, the holder list would contain only one entry. Since
a time (file locking). While safe, this method can cause this type of lock effectively blocks any other entity that
unnecessary waiting.
requires the lock from processing, care must be used to:
If the clerks can remove one page from the ledger, containing the account of the current customer (plus several
other accounts), then multiple customers can be serviced
concurrently, provided that each customer’s account is
found on a different page than the others. If two customers have accounts on the same page, then only one
may be serviced at a time. This is analogous to a page
level lock in a database.
A higher degree of granularity is achieved if each individual account may be taken by a clerk. This would allow
any customer to be serviced without waiting for another
customer who is accessing a different account. This is
analogous to a record level lock and is normally the highest
degree of locking granularity in a database management
In a SQL database, a record is typically called a “row.”
The introduction of granular (subset) locks creates the
possibility for a situation called deadlock. Deadlock is
possible when incremental locking (locking one entity,
then locking one or more additional entities) is used. To
illustrate, if two bank customers asked two clerks to obtain their account information so they could transfer some
money into other accounts, the two accounts would essentially be locked. Then, if the customers told their clerks
that the money was to be transferred into each other’s accounts, the clerks would search for the other accounts but
find them to be “in use” and wait for them to be returned.
Unknowingly, the two clerks are waiting for each other,
and neither of them can complete their transaction until
the other gives up and returns the account. Various techniques are used to avoid such problems.
Use of locks
• ensure the lock is held for the shortest time possible;
• not hold the lock across system or function calls
where the entity is no longer running on the processor - this can lead to deadlock;
• ensure that if the entity is unexpectedly exited for
any reason, the lock is freed.
Non-holders of the lock (aka waiters) can be held in a
list that is serviced in a round robin fashion, or in a FIFO
queue. This would ensure that any possible waiter would
get equal chance to obtain the lock and not be locked out.
To further speed up the process, if an entity has gone to
sleep waiting for a lock, performance is improved if the
entity is notified of the grant, instead of discovering it on
some sort of system timeout driven wakeup.
Shared locks
Shared locks differ from exclusive locks in that the
holder list can contain multiple entries. Shared locks allow all holders to read the contents of the record knowing
that the record cannot be changed until after the lock has
been released by all holders. Exclusive locks cannot be
obtained when a record is already locked (exclusively or
shared) by another entity.
If lock requests for the same entity are queued, then once
a shared lock is granted, any queued shared locks may
also be granted. If an exclusive lock is found next on the
queue, it must wait until all shared locks have been released. As with exclusive locks, these shared locks should
be held for the least time possible.
Record locks need to be managed between the entities re- 6.2.3 References
questing the records such that no entity is given too much
service via successive grants, and no other entity is ef- [1] Gray, Jim, and Reuter, Andreas (1993), Distributed
Transaction Processing: Concepts and Techniques, Morfectively locked out. The entities that request a lock can
gan Kaufmann, pp. 375–437, ISBN 1-55860-190-2
be either individual applications (programs) or an entire
The application or system should be designed such that
any lock is held for the shortest time possible. Data read- 6.3 Two-phase locking
ing, without editing facilities, does not require a lock, and
This article is about concurrency control. For commit
reading locked records is usually permissible.
consensus within a distributed transaction, see TwoTwo main types of locks can be requested:
phase commit protocol.
Exclusive locks
In databases and transaction processing, two-phase locking (2PL) is a concurrency control method that guaranExclusive locks are, as the name implies, exclusively held tees serializability.[1][2] It is also the name of the resulting
by a single entity, usually for the purpose of writing to set of database transaction schedules (histories). The prothe record. If the locking schema was represented by a tocol utilizes locks, applied by a transaction to data, which
may block (interpreted as signals to stop) other transactions from accessing the same data during the transaction’s life.
By the 2PL protocol locks are applied and removed in
two phases:
1. Expanding phase: locks are acquired and no locks
are released.
2. Shrinking phase: locks are released and no locks are
Two types of locks are utilized by the basic protocol:
Shared and Exclusive locks. Refinements of the basic protocol may utilize more lock types. Using locks that block
processes, 2PL may be subject to deadlocks that result
from the mutual blocking of two or more transactions.
Data-access locks
A lock is a system object associated with a shared resource such as a data item of an elementary type, a row in
a database, or a page of memory. In a database, a lock on
a database object (a data-access lock) may need to be acquired by a transaction before accessing the object. Correct use of locks prevents undesired, incorrect or inconsistent operations on shared resources by other concurrent
transactions. When a database object with an existing
lock acquired by one transaction needs to be accessed by
another transaction, the existing lock for the object and
the type of the intended access are checked by the system.
If the existing lock type does not allow this specific attempted concurrent access type, the transaction attempting access is blocked (according to a predefined agreement/scheme). In practice a lock on an object does not
directly block a transaction’s operation upon the object,
but rather blocks that transaction from acquiring another
lock on the same object, needed to be held/owned by the
transaction before performing this operation. Thus, with
a locking mechanism, needed operation blocking is controlled by a proper lock blocking scheme, which indicates
which lock type blocks which lock type.
Two major types of locks are utilized:
• An existing write-lock on a database object blocks
an intended write upon the same object (already requested/issued) by another transaction by blocking
a respective write-lock from being acquired by the
other transaction. The second write-lock will be acquired and the requested write of the object will take
place (materialize) after the existing write-lock is
• A write-lock blocks an intended (already requested/issued) read by another transaction by
blocking the respective read-lock .
• A read-lock blocks an intended write by another
transaction by blocking the respective write-lock .
• A read-lock does not block an intended read by
another transaction. The respective read-lock for
the intended read is acquired (shared with the previous read) immediately after the intended read is
requested, and then the intended read itself takes
Several variations and refinements of these major lock
types exist, with respective variations of blocking behavior. If a first lock blocks another lock, the two locks are
called incompatible; otherwise the locks are compatible.
Often lock types blocking interactions are presented in
the technical literature by a Lock compatibility table. The
following is an example with the common, major lock
X indicates incompatibility, i.e, a case when a
lock of the first type (in left column) on an object blocks a lock of the second type (in top
row) from being acquired on the same object
(by another transaction). An object typically
has a queue of waiting requested (by transactions) operations with respective locks. The
first blocked lock for operation in the queue is
acquired as soon as the existing blocking lock
is removed from the object, and then its respective operation is executed. If a lock for operation in the queue is not blocked by any existing
lock (existence of multiple compatible locks on
a same object is possible concurrently) it is acquired immediately.
Comment: In some publications the table en• Write-lock (exclusive lock) is associated with a
tries are simply marked “compatible” or “indatabase object by a transaction (Terminology: “the
compatible”, or respectively “yes” or “no”.
transaction locks the object,” or “acquires lock
for it”) before writing (inserting/modifying/deleting)
6.3.2 Two-phase locking and its special
this object.
• Read-lock (shared lock) is associated with a
database object by a transaction before reading (re- Two-phase locking
trieving the state of) this object.
According to the two-phase locking protocol a transacThe common interactions between these lock types are tion handles its locks in two distinct, consecutive phases
during the transaction’s execution:
defined by blocking behavior as follows:
1. Expanding phase (aka Growing phase): locks are SS2PL has been the concurrency control protocol of
acquired and no locks are released (the number of choice for most database systems and utilized since their
locks can only increase).
early days in the 1970s. It is proven to be an effective mechanism in many situations, and provides besides
2. Shrinking phase: locks are released and no locks Serializability also Strictness (a special case of cascadeare acquired.
less Recoverability), which is instrumental for efficient
database recovery, and also Commitment ordering (CO)
The two phase locking rule can be summarized as: never for participating in distributed environments where a CO
acquire a lock after a lock has been released. The based distributed serializability and global serializability
serializability property is guaranteed for a schedule with solutions are employed. Being a subset of CO, an efficient implementation of distributed SS2PL exists withtransactions that obey this rule.
out a distributed lock manager (DLM), while distributed
Typically, without explicit knowledge in a transaction on deadlocks (see below) are resolved automatically. The
end of phase-1, it is safely determined only when a trans- fact that SS2PL employed in multi database systems enaction has completed processing and requested commit. sures global serializability has been known for years beIn this case all the locks can be released at once (phase-2). fore the discovery of CO, but only with CO came the understanding of the role of an atomic commitment protocol in maintaining global serializability, as well as the
Strict two-phase locking
observation of automatic distributed deadlock resolution
(see a detailed example of Distributed SS2PL). As a matTo comply with the S2PL protocol a transaction needs to ter of fact, SS2PL inheriting properties of Recoverabilcomply with 2PL, and release its write (exclusive) locks ity and CO is more significant than being a subset of
only after it has ended, i.e., being either committed or 2PL, which by itself in its general form, besides comprisaborted. On the other hand, read (shared) locks are re- ing a simple serializability mechanism (however serializleased regularly during phase 2. This protocol is not ap- ability is also implied by CO), in not known to provide
propriate in B-trees because it causes Bottleneck (while SS2PL with any other significant qualities. 2PL in its
B-trees always starts searching from the parent root).
general form, as well as when combined with Strictness,
i.e., Strict 2PL (S2PL), are not known to be utilized in
practice. The popular SS2PL does not require marking
Strong strict two-phase locking
“end of phase-1” as 2PL and S2PL do, and thus is simpler to implement. Also, unlike the general 2PL, SS2PL
or Rigorousness, or Rigorous scheduling, or Rigorous provides, as mentioned above, the useful Strictness and
two-phase locking
Commitment ordering properties.
To comply with strong strict two-phase locking Many variants of SS2PL exist that utilize various lock
(SS2PL) the locking protocol releases both write (exclu- types with various semantics in different situations, insive) and read (shared) locks applied by a transaction only cluding cases of lock-type change during a transaction.
after the transaction has ended, i.e., only after both com- Notable are variants that use Multiple granularity lockpleting executing (being ready) and becoming either com- ing.
mitted or aborted. This protocol also complies with the
S2PL rules. A transaction obeying SS2PL can be viewed Comments:
as having phase-1 that lasts the transaction’s entire execu1. SS2PL Vs. S2PL: Both provide Serializability
tion duration, and no phase-2 (or a degenerate phase-2).
and Strictness. Since S2PL is a super class of
Thus, only one phase is actually left, and “two-phase” in
SS2PL it may, in principle, provide more concurthe name seems to be still utilized due to the historical
rency. However, no concurrency advantage is typidevelopment of the concept from 2PL, and 2PL being a
cally practically noticed (exactly same locking exists
super-class. The SS2PL property of a schedule is also
for both, with practically not much earlier lock recalled Rigorousness. It is also the name of the class of
lease for S2PL), and the overhead of dealing with
schedules having this property, and an SS2PL schedule
an end-of-phase-1 mechanism in S2PL, separate
is also called a “rigorous schedule”. The term “Rigorousfrom transaction-end, is not justified. Also, while
ness” is free of the unnecessary legacy of “two-phase,”
SS2PL provides Commitment ordering, S2PL does
as well as being independent of any (locking) mechanism
not. This explains the preference of SS2PL over
(in principle other blocking mechanisms can be utilized).
The property’s respective locking mechanism is sometimes referred to as Rigorous 2PL.
2. Especially before 1990, but also after, in many arSS2PL is a special case of S2PL, i.e., the SS2PL class
of schedules is a proper subclass of S2PL (every SS2PL
schedule is also an S2PL schedule, but S2PL schedules
exist that are not SS2PL).
ticles and books, e.g., (Bernstein et al. 1987, p.
59),[1] the term “Strict 2PL” (S2PL) has been frequently defined by the locking protocol “Release all
locks only after transaction end,” which is the pro-
tocol of SS2PL. Thus, “Strict 2PL” could not be
there the name of the intersection of Strictness and
2PL, which is larger than the class generated by the
SS2PL protocol. This has caused confusion. With
an explicit definition of S2PL as the intersection of
Strictness and 2PL, a new name for SS2PL, and an
explicit distinction between the classes S2PL and
SS2PL, the articles (Breitbart et al. 1991)[3] and
(Raz 1992)[4] have intended to clear the confusion:
The first using the name “Rigorousness,” and the
second “SS2PL.”
3. A more general property than SS2PL exists (a
schedule super-class), Strict commitment ordering (Strict CO, or SCO), which as well provides
both serializability, strictness, and CO, and has similar locking overhead. Unlike SS2PL, SCO does
not block upon a read-write conflict (a read-lock
does not block acquiring a write-lock; both SCO and
SS2PL have the same behavior for write-read and
write-write conflicts) at the cost of a possible delayed commit, and upon such conflict type SCO has
shorter average transaction completion time and better performance than SS2PL.[5] While SS2PL obeys Schedule classes containment: An arrow from class A to class
the lock compatibility table above, SCO has the fol- B indicates that class A strictly contains B; a lack of a directed
path between classes means that the classes are incomparable.
lowing table:
Note that though SCO releases all
locks at transaction end and complies with the 2PL locking rules,
SCO is not a subset of 2PL because of its different lock compatibility table. SCO allows materialized read-write conflicts between
two transactions in their phases 1,
which 2PL does not allow in phase1 (see about materialized conflicts
in Serializability). On the other
hand 2PL allows other materialized conflict types in phase-2 that
SCO does not allow at all. Together this implies that the schedule
classes 2PL and SCO are incomparable (i.e., no class contains the
other class).
Summary - Relationships among classes
A property is inherently blocking, if it can be enforced only by
blocking transaction’s data access operations until certain events
occur in other transactions. (Raz 1992)
refers to a different mechanism with a class that includes
also schedules not in the 2PL class).
6.3.3 Deadlocks in 2PL
Locks block data-access operations. Mutual blocking between transactions results in a deadlock, where execution
of these transactions is stalled, and no completion can be
reached. Thus deadlocks need to be resolved to complete
these transactions’ executions and release related computing resources. A deadlock is a reflection of a potential cycle in the precedence graph, that would occur without the
blocking. A deadlock is resolved by aborting a transaction involved with such potential cycle, and breaking the
cycle. It is often detected using a wait-for graph (a graph
of conflicts blocked by locks from being materialized;
conflicts not materialized in the database due to blocked
operations are not reflected in the precedence graph and
do not affect serializability), which indicates which transaction is “waiting for” lock release by which transaction,
and a cycle means a deadlock. Aborting one transaction
per cycle is sufficient to break the cycle. Transactions
aborted due to deadlock resolution are executed again immediately.
Between any two schedule classes (define by their schedules’ respective properties) that have common schedules,
either one contains the other (strictly contains if they are
not equal), or they are incomparable. The containment
relationships among the 2PL classes and other major
schedule classes are summarized in the following diagram. 2PL and its subclasses are inherently blocking, In a distributed environment an atomic commitment prowhich means that no optimistic implementations for them tocol, typically the Two-phase commit (2PC) protocol, is
exist (and whenever “Optimistic 2PL” is mentioned it utilized for atomicity. When recoverable data (data under
transaction control) are partitioned among 2PC participants (i.e., each data object is controlled by a single 2PC
participant), then distributed (global) deadlocks, deadlocks involving two or more participants in 2PC, are resolved automatically as follows:
vote handling upon voting
deadlock, resulting in global
deadlock resolution) has not
been mentioned until today
(2009). Practically only the
special case SS2PL is utilized,
where no end-of-phase-one
synchronization is needed in
addition to atomic commit
When SS2PL is effectively utilized in a distributed environment, then global deadlocks due to locking generate voting-deadlocks in 2PC, and are resolved automatically by 2PC (see Commitment ordering (CO), in Exact
characterization of voting-deadlocks by global cycles; No
reference except the CO articles is known to notice this).
In a distributed environment where recoverFor the general case of 2PL, global deadlocks are simiable data are not partitioned among atomic
larly resolved automatically by the synchronization point
commitment protocol participants, no such auprotocol of phase-1 end in a distributed transaction (syntomatic resolution exists, and distributed deadchronization point is achieved by “voting” (notifying local
locks need to be resolved by dedicated techphase-1 end), and being propagated to the participants in
a distributed transaction the same way as a decision point
in atomic commitment; in analogy to decision point in
CO, a conflicting operation in 2PL cannot happen before 6.3.4 See also
phase-1 end synchronization point, with the same result• Serializability
ing voting-deadlock in the case of a global data-access
deadlock; the voting-deadlock (which is also a locking
• Lock (computer science)
based global deadlock) is automatically resolved by the
protocol aborting some transaction involved, with a missing vote, typically using a timeout).
6.3.5 References
When data are partitioned among the atomic
commitment protocol (e.g., 2PC) participants,
automatic global deadlock resolution has been
overlooked in the database research literature,
though deadlocks in such systems has been a
quite intensive research area:
• For CO and its special case
SS2PL, the automatic resolution by the atomic commitment
protocol has been noticed only
in the CO articles. However,
it has been noticed in practice
that in many cases global deadlocks are very infrequently detected by the dedicated resolution mechanisms, less than
could be expected (“Why do
we see so few global deadlocks?"). The reason is probably the deadlocks that are automatically resolved and thus not
handled and uncounted by the
• For 2PL in general, the
automatic resolution by the
(mandatory) end-of-phase-one
synchronization point protocol
(which has same voting mechanism as atomic commitment
protocol, and same missing
[1] Philip A. Bernstein, Vassos Hadzilacos, Nathan Goodman
(1987): Concurrency Control and Recovery in Database
Systems, Addison Wesley Publishing Company, ISBN 0201-10715-5
[2] Gerhard Weikum,
Gottfried Vossen (2001):
Transactional Information Systems, Elsevier, ISBN
[3] Yuri Breitbart, Dimitrios Georgakopoulos, Marek
Rusinkiewicz, Abraham Silberschatz (1991): “On
Rigorous Transaction Scheduling”, IEEE Transactions on
Software Engineering (TSE), September 1991, Volume
17, Issue 9, pp. 954-960, ISSN: 0098-5589
[4] Yoav Raz (1992): “The Principle of Commitment Ordering, or Guaranteeing Serializability in a Heterogeneous
Environment of Multiple Autonomous Resource Managers Using Atomic Commitment” (PDF), Proceedings
of the Eighteenth International Conference on Very Large
Data Bases (VLDB), pp. 292-312, Vancouver, Canada,
August 1992, ISBN 1-55860-151-1 (also DEC-TR 841,
Digital Equipment Corporation, November 1990)
[5] Yoav Raz (1991): “Locking Based Strict Commitment
Ordering, or How to improve Concurrency in Locking
Based Resource Managers”, DEC-TR 844, December
Chapter 7
7.1 Multiversion concurrency control
rent reads access the older version.
Multiversion concurrency control (MCC or MVCC),
is a concurrency control method commonly used by 7.1.1 Implementation
database management systems to provide concurrent access to the database and in programming languages to im- MVCC uses timestamps (TS), and incrementing transacplement transactional memory.[1]
tion IDs (T), to achieve transactional consistency. MVCC
If someone is reading from a database at the same time ensures a transaction (T) never has to wait to Read a
as someone else is writing to it, it is possible that the database object (P) by maintaining several versions of
reader will see a half-written or inconsistent piece of data. such object (P). Each version of the object (P) would
There are several ways of solving this problem, known have both a Read Timestamp (RTS) and a Write Timesas concurrency control methods. The simplest way is to tamp (WTS) which lets a transaction (Tᵢ) read the most
make all readers wait until the writer is done, which is recent version of an object (P) which precedes the transknown as a lock. This can be very slow, so MVCC takes action’s (Tᵢ) Read Timestamp (RTS(Tᵢ)).
a different approach: each user connected to the database
sees a snapshot of the database at a particular instant in
time. Any changes made by a writer will not be seen by
other users of the database until the changes have been
completed (or, in database terms: until the transaction
has been committed.)
When an MVCC database needs to update an item of
data, it will not overwrite the old data with new data,
but instead mark the old data as obsolete and add the
newer version elsewhere. Thus there are multiple versions
stored, but only one is the latest. This allows readers to access the data that was there when they began reading, even
if it was modified or deleted part way through by someone
else. It also allows the database to avoid the overhead of
filling in holes in memory or disk structures but requires
(generally) the system to periodically sweep through and
delete the old, obsolete data objects. For a documentoriented database it also allows the system to optimize
documents by writing entire documents onto contiguous
sections of disk—when updated, the entire document can
be re-written rather than bits and pieces cut out or maintained in a linked, non-contiguous database structure.
MVCC provides point in time consistent views. Read
transactions under MVCC typically use a timestamp or
transaction ID to determine what state of the DB to read,
and read these versions of the data. Read and write transactions are thus isolated from each other without any need
for locking. Writes create a newer version, while concur-
If a transaction (Tᵢ) wants to Write to an object (P), and
if there is also another transaction (T ) happening to the
same object (P), the Read Timestamp (RTS) of (Tᵢ)
must precede the Read Timestamp (RTS) of (T ), (i.e.,
RTS(Tᵢ) < RTS(T )) for the object (P) Write Operation
(WTS) to succeed. Basically, a Write cannot complete
if there are other outstanding transactions with an earlier
Read Timestamp (RTS) to the same object (P). Think of
it like standing in line at the store, you cannot complete
your checkout transaction until those in front of you have
completed theirs.
To restate; every object (P) has a Timestamp (TS), however if transaction Tᵢ wants to Write to object (P), and
there is a Timestamp (TS) of that transaction that is earlier than the object’s current Read Timestamp, (TS(P) <
RTS(Tᵢ)), the transaction Tᵢ is aborted and restarted. (If
you try to cut in line, to check out early, go to the back of
that line) Otherwise, Tᵢ creates a new version of (P) and
sets the read/write timestamp (TS) of the new version of
(P) to the timestamp of the transaction TS=TS(Tᵢ).[2]
The obvious drawback to this system is the cost of storing
multiple versions of objects in the database. On the other
hand reads are never blocked, which can be important
for workloads mostly involving reading values from the
database. MVCC is particularly adept at implementing
true snapshot isolation, something which other methods
of concurrency control frequently do either incompletely
or with high performance costs.
Concurrent read-write
At Time = 1, the state of a database could be:
T0 wrote Object 1="Foo” and Object 2="Bar”. After
that T1 wrote Object 1="Hello” leaving Object 2 at its
original value. The new value of Object 1 will supersede
the value at 0 for all transaction that starts after T1 commits at which point version 0 of Object 1 can be garbage
If a long running transaction T2 starts a read operation of
Object 2 and Object 1 after T1 committed and there is
a concurrent update transaction T3 which deletes Object
2 and adds Object 3="Foo-Bar”, the database state will
look like at time 2:
There is a new version as of time 2 of Object 2 which
is marked as deleted and a new Object 3. Since T2 and
T3 run concurrently T2 sees another the version of the
database before 2 i.e. before T3 committed writes, as
such T2 reads Object 2="Bar” and Object 1="Hello”.
This is how MVCC allows snapshot isolation reads in almost every case without any locks.
7.1.5 See also
• List of databases using MVCC
• Timestamp-based concurrency control
• Clojure
• Read-copy-update
• Vector clock
7.1.6 References
[1] refs. Clojure. Retrieved on 2013-09-18.
[2] Ramakrishnan, R., & Gehrke, J. (2000). Database management systems. Osborne/McGraw-Hill.
[3] Bernstein, Philip A.; Goodman, Nathan (1981).
“Concurrency Control in Distributed Database Systems”.
ACM Computing Surveys.
[4] Reed, David P. (September 21, 1978). “Naming and Synchronization in a Decentralized Computer System”. MIT
7.1.7 Further reading
• Gerhard Weikum, Gottfried Vossen, Transactional
information systems: theory, algorithms, and the
practice of concurrency control and recovery, Morgan Kaufmann, 2002, ISBN 1-55860-508-8
Multiversion concurrency control is described in some
detail in the 1981 paper “Concurrency Control in Distributed Database Systems”[3] by Phil Bernstein and
Nathan Goodman, then employed by the Computer Corporation of America. Bernstein and Goodman’s paper 7.2 Snapshot isolation
cites a 1978 dissertation[4] by David P. Reed which quite
clearly describes MVCC and claims it as an original work. In databases, and transaction processing (transaction
management), snapshot isolation is a guarantee that all
The first shipping, commercial database software product
reads made in a transaction will see a consistent snapshot
featuring MVCC was Digital’s VAX Rdb/ELN. The secof the database (in practice it reads the last committed
ond was InterBase, which is still an active, commercial
values that existed at the time it started), and the transacproduct.
tion itself will successfully commit only if no updates it
has made conflict with any concurrent updates made since
that snapshot.
Version control systems
Any version control system that has the internal notion of
a version (e.g. Subversion, Git, probably almost any current VCS with the notable exception of CVS) will provide
explicit MVCC (you only ever access data by its version
Among the VCSs that don't provide MVCC at the repository level, most still work with the notion of a working
copy, which is a file tree checked out from the repository,
edited without using the VCS itself and checked in after
the edit. This working copy provides MVCC while it is
checked out.
Snapshot isolation has been adopted by several major
database management systems, such as SQL Anywhere,
InterBase, Firebird, Oracle, PostgreSQL, MongoDB[1]
and Microsoft SQL Server (2005 and later). The main
reason for its adoption is that it allows better performance
than serializability, yet still avoids most of the concurrency anomalies that serializability avoids (but not always
all). In practice snapshot isolation is implemented within
multiversion concurrency control (MVCC), where generational values of each data item (versions) are maintained:
MVCC is a common way to increase concurrency and
performance by generating a new version of a database
object each time the object is written, and allowing transactions’ read operations of several last relevant versions
(of each object). Snapshot isolation has also been used[2]
to critique the ANSI SQL−92 standard’s definition of
isolation levels, as it exhibits none of the “anomalies” that
the SQL standard prohibited, yet is not serializable (the
anomaly-free isolation level defined by ANSI).
If built on multiversion concurrency control, snapshot
isolation allows transactions to proceed without worrying about concurrent operations, and more importantly
without needing to re-verify all read operations when the
transaction finally commits. The only information that
Snapshot isolation is called “serializable” mode in must be stored during the transaction is a list of updates
Oracle[3][4] <ref name"">Ask Tom : “Serializable made, which can be scanned for conflicts fairly easily beTransaction”</ref> and PostgreSQL versions prior to fore being committed.
9.1,[5][6][7] which may cause confusion with the “real
serializability" mode. There are arguments both for and 7.2.2 Workarounds
against this decision; what is clear is that users must
be aware of the distinction to avoid possible undesired Potential inconsistency problems arising from write skew
anomalous behavior in their database system logic.
anomalies can be fixed by adding (otherwise unnecessary) updates to the transactions in order to enforce the
serializability property.[8]
Materialize the conflict Add a special conflict table,
A transaction executing under snapshot isolation appears
which both transactions update in order to create a
to operate on a personal snapshot of the database, taken
direct write-write conflict.
at the start of the transaction. When the transaction concludes, it will successfully commit only if the values up- Promotion Have one transaction “update” a read-only
location (replacing a value with the same value) in
dated by the transaction have not been changed externally
order to create a direct write-write conflict (or use an
since the snapshot was taken. Such a write-write conflict
equivalent promotion, e.g. Oracle’s SELECT FOR
will cause the transaction to abort.
In a write skew anomaly, two transactions (T1 and T2)
concurrently read an overlapping data set (e.g. values V1
and V2), concurrently make disjoint updates (e.g. T1 up- In the example above, we can materialize the conflict by
dates V1, T2 updates V2), and finally concurrently com- adding a new table which makes the hidden constraint exmit, neither having seen the update performed by the plicit, mapping each person to their total balance. Phil
other. Were the system serializable, such an anomaly would start off with a total balance of $200, and each
would be impossible, as either T1 or T2 would have to transaction would attempt to subtract $200 from this, creoccur “first”, and be visible to the other. In contrast, snap- ating a write-write conflict that would prevent the two
from succeeding concurrently. This approach violates the
shot isolation permits write skew anomalies.
normal form.
As a concrete example, imagine V1 and V2 are two balances held by a single person, Phil. The bank will allow Alternatively, we can promote one of the transaction’s
either V1 or V2 to run a deficit, provided the total held in reads to a write. For instance, T2 could set V1 = V1, creboth is never negative (i.e. V1 + V2 ≥ 0). Both balances ating an artificial write-write conflict with T1 and, again,
are currently $100. Phil initiates two transactions con- preventing the two from succeeding concurrently. This
currently, T1 withdrawing $200 from V1, and T2 with- solution may not always be possible.
drawing $200 from V2.
In general, therefore, snapshot isolation puts some of the
If the database guaranteed serializable transactions, the problem of maintaining non-trivial constraints onto the
simplest way of coding T1 is to deduct $200 from V1, user, who may not appreciate either the potential pitfalls
and then verify that V1 + V2 ≥ 0 still holds, aborting if or the possible solutions. The upside to this transfer is
not. T2 similarly deducts $200 from V2 and then verifies better performance.
V1 + V2 ≥ 0. Since the transactions must serialize, either
T1 happens first, leaving V1 = -$100, V2 = $100, and
7.2.3 History
preventing T2 from succeeding (since V1 + (V2 - $200)
is now -$200), or T2 happens first and similarly prevents Snapshot isolation arose from work on multiversion conT1 from committing.
currency control databases, where multiple versions of
Under snapshot isolation, however, T1 and T2 operate the database are maintained concurrently to allow readers
on private snapshots of the database: each deducts $200 to execute without colliding with writers. Such a system
from an account, and then verifies that the new total is allows a natural definition and implementation of such an
zero, using the other account value that held when the isolation level.[2] InterBase, later owned by Borland, was
snapshot was taken. Since neither update conflicts, both acknowledged to provide SI rather than full serializability
commit successfully, leaving V1 = V2 = -$100, and V1 in version 4,[2] and likely permitted write-skew anomalies
since its first release in 1985.[9]
+ V2 = -$200.
Unfortunately, the ANSI SQL-92 standard was written 7.2.5 Further reading
with a lock-based database in mind, and hence is rather
vague when applied to MVCC systems. Berenson et al.
• Gerhard Weikum, Gottfried Vossen, Transactional
wrote a paper in 1995[2] critiquing the SQL standard, and
information systems: theory, algorithms, and the
cited snapshot isolation as an example of an isolation level
practice of concurrency control and recovery, Morthat did not exhibit the standard anomalies described in
gan Kaufmann, 2002, ISBN 1-55860-508-8
the ANSI SQL-92 standard, yet still had anomalous behaviour when compared with serializable transactions.
• Khuzaima Daudjee, Kenneth Salem, Lazy Database
Replication with Snapshot Isolation, VLDB 2006:
In 2008, Cahill et al. showed that write-skew anomalies
pages 715-726
could be prevented by detecting and aborting “danger[10]
ous” triplets of concurrent transactions.
This implementation of serializability is well-suited to multiversion
concurrency control databases, and has been adopted in 7.3 Two-phase commit protocol
PostgreSQL 9.1,[6][7][11] where it is referred to as “Serializable Snapshot Isolation”, abbreviated to SSI. When
used consistently, this eliminates the need for the above “2PC” redirects here. For the play in American and
workarounds. The downside over snapshot isolation is Canadian football, see Two-point conversion. For the
an increase in aborted transactions. This can perform American rapper, see 2Pac. For the cryptographic
better or worse than snapshot isolation with the above protocol, see Commitment scheme.
workarounds, depending on workload.
In transaction processing, databases, and computer networking, the two-phase commit protocol (2PC) is a
7.2.4 References
type of atomic commitment protocol (ACP). It is a
[1] Multiversion concurrency control in MongoDB, distributed algorithm that coordinates all the processes
MongoDB CTO: How our new WiredTiger storage that participate in a distributed atomic transaction on
whether to commit or abort (roll back) the transaction (it
engine will earn its stripes
is a specialized type of consensus protocol). The protocol
[2] Berenson, Hal; Bernstein, Phil; Gray, Jim; Melton, Jim;
achieves its goal even in many cases of temporary system
O'Neil, Elizabeth; O'Neil, Patrick (1995), “A Critique
failure (involving either process, network node, commuof ANSI SQL Isolation Levels”, Proceedings of the 1995
ACM SIGMOD international Conference on Management nication, etc. failures), and is thus widely utilized.
configuof Data, pp. 1–10, doi:10.1145/223784.223785
rations, and in rare cases, user (e.g., a system’s adminis[3] Oracle Database Concepts 10g Release 1 (10.1) Chapter trator) intervention is needed to remedy an outcome. To
13 : Data Concurrency and Consistency — Oracle Isolaaccommodate recovery from failure (automatic in most
tion Levels
cases) the protocol’s participants use logging of the pro[4] Ask Tom : On Transaction Isolation Levels
tocol’s states. Log records, which are typically slow to
generate but survive failures, are used by the protocol’s
[5] PostgreSQL 9.0 Documentation: Serializable
recovery procedures. Many protocol variants exist that
Isolation versus True Serializability
primarily differ in logging strategies and recovery mech[6] PostgreSQL 9.1 press release
anisms. Though usually intended to be used infrequently,
[7] PostgreSQL 9.1.14 Documentation: 13.2.3. Serializable recovery procedures compose a substantial portion of the
protocol, due to many possible failure scenarios to be
Isolation Level
considered and supported by the protocol.
[8] Fekete, Alan; Liarokapis, Dimitrios; O'Neil, Elizabeth; O'Neil, Patrick; Shasha, Dennis (2005),
“Making Snapshot Isolation Serializable”, ACM
Transactions on Database Systems 30 (2): 492–528,
doi:10.1145/1071610.1071615, ISSN 0362-5915
[9] Stuntz, Craig. “Multiversion Concurrency Control Before
InterBase”. Retrieved October 30, 2014.
[10] Michael J. Cahill, Uwe Röhm, Alan D. Fekete (2008)
“Serializable isolation for snapshot databases”, Proceedings of the 2008 ACM SIGMOD international conference
on Management of data, pp. 729–738, ISBN 978-160558-102-6 (SIGMOD 2008 best paper award)
[11] Ports, Dan R. K.; Grittner, Kevin (2012). “Serializable
Snapshot Isolation in PostgreSQL” (PDF). Proceedings of
the VLDB Endowment 5 (12): 1850–1861.
In a “normal execution” of any single distributed transaction, i.e., when no failure occurs, which is typically
the most frequent situation, the protocol consists of two
1. The commit-request phase (or voting phase), in
which a coordinator process attempts to prepare
all the transaction’s participating processes (named
participants, cohorts, or workers) to take the necessary steps for either committing or aborting the
transaction and to vote, either “Yes": commit (if the
transaction participant’s local portion execution has
ended properly), or “No": abort (if a problem has
been detected with the local portion), and
2. The commit phase, in which, based on voting of the Success If the coordinator received an agreement
cohorts, the coordinator decides whether to com- message from all cohorts during the commit-request
mit (only if all have voted “Yes”) or abort the trans- phase:
action (otherwise), and notifies the result to all the
cohorts. The cohorts then follow with the needed
1. The coordinator sends a commit message to all the
actions (commit or abort) with their local transaccohorts.
tional resources (also called recoverable resources;
e.g., database data) and their respective portions in
2. Each cohort completes the operation, and releases
the transaction’s other output (if applicable).
all the locks and resources held during the transaction.
Note that the two-phase commit (2PC) protocol should
not be confused with the two-phase locking (2PL) proto3. Each cohort sends an acknowledgment to the coorcol, a concurrency control protocol.
4. The coordinator completes the transaction when all
acknowledgments have been received.
The protocol works in the following manner: one node is
designated the coordinator, which is the master site, and Failure If any cohort votes No during the committhe rest of the nodes in the network are designated the co- request phase (or the coordinator’s timeout expires):
horts. The protocol assumes that there is stable storage
at each node with a write-ahead log, that no node crashes
1. The coordinator sends a rollback message to all the
forever, that the data in the write-ahead log is never lost
or corrupted in a crash, and that any two nodes can communicate with each other. The last assumption is not too
2. Each cohort undoes the transaction using the undo
restrictive, as network communication can typically be
log, and releases the resources and locks held during
rerouted. The first two assumptions are much stronger; if
the transaction.
a node is totally destroyed then data can be lost.
The protocol is initiated by the coordinator after the last
step of the transaction has been reached. The cohorts then
respond with an agreement message or an abort message
depending on whether the transaction has been processed
successfully at the cohort.
Basic algorithm
Commit request phase
or voting phase
3. Each cohort sends an acknowledgement to the coordinator.
4. The coordinator undoes the transaction when all acknowledgements have been received.
Message flow Coordinator Cohort QUERY TO
COMMIT --------------------------------> VOTE YES/NO
prepare*/abort* <------------------------------- commit*/abort* COMMIT/ROLLBACK -------------------------------> ACKNOWLEDGMENT commit*/abort*
<-------------------------------- end
1. The coordinator sends a query to commit message
An * next to the record type means that the record is
to all cohorts and waits until it has received a reply
forced to stable storage.[4]
from all cohorts.
2. The cohorts execute the transaction up to the point
where they will be asked to commit. They each write 7.3.3 Disadvantages
an entry to their undo log and an entry to their redo
The greatest disadvantage of the two-phase commit protocol is that it is a blocking protocol. If the coordinator
3. Each cohort replies with an agreement message
fails permanently, some cohorts will never resolve their
(cohort votes Yes to commit), if the cohort’s actions
transactions: After a cohort has sent an agreement messucceeded, or an abort message (cohort votes No,
sage to the coordinator, it will block until a commit or
not to commit), if the cohort experiences a failure
rollback is received.
that will make it impossible to commit.
Commit phase
or Completion phase
7.3.4 Implementing the two-phase commit
Common architecture
type. Thus the best variant of optimization, if any, is
chosen according to failure and transaction outcome
In many cases the 2PC protocol is distributed in a com- statistics.
puter network. It is easily distributed by implementing multiple dedicated 2PC components similar to each
other, typically named Transaction managers (TMs; also Tree two-phase commit protocol The Tree 2PC proreferred to as 2PC agents or Transaction Processing Mon- tocol[2] (also called Nested 2PC, or Recursive 2PC) is a
itors), that carry out the protocol’s execution for each common variant of 2PC in a computer network, which
transaction (e.g., The Open Group's X/Open XA). The better utilizes the underlying communication infrastrucdatabases involved with a distributed transaction, the par- ture. The participants in a distributed transaction are
ticipants, both the coordinator and cohorts, register to typically invoked in an order which defines a tree strucclose TMs (typically residing on respective same network ture, the invocation tree, where the participants are the
nodes as the participants) for terminating that transaction nodes and the edges are the invocations (communication
using 2PC. Each distributed transaction has an ad hoc set links). The same tree is commonly utilized to complete
of TMs, the TMs to which the transaction participants the transaction by a 2PC protocol, but also another comregister. A leader, the coordinator TM, exists for each munication tree can be utilized for this, in principle. In a
transaction to coordinate 2PC for it, typically the TM of tree 2PC the coordinator is considered the root (“top”) of
the coordinator database. However, the coordinator role a communication tree (inverted tree), while the cohorts
can be transferred to another TM for performance or re- are the other nodes. The coordinator can be the node
liability reasons. Rather than exchanging 2PC messages that originated the transaction (invoked recursively (tranamong themselves, the participants exchange the mes- sitively) the other participants), but also another node in
sages with their respective TMs. The relevant TMs com- the same tree can take the coordinator role instead. 2PC
municate among themselves to execute the 2PC protocol messages from the coordinator are propagated “down”
schema above, “representing” the respective participants, the tree, while messages to the coordinator are “collected”
for terminating that transaction. With this architecture by a cohort from all the cohorts below it, before it sends
the protocol is fully distributed (does not need any cen- the appropriate message “up” the tree (except an abort
tral processing component or data structure), and scales message, which is propagated “up” immediately upon reup with number of network nodes (network size) effec- ceiving it or if the current cohort initiates the abort).
The Dynamic two-phase commit (Dynamic two-phase
This common architecture is also effective for the distribution of other atomic commitment protocols besides 2PC, since all such protocols use the same voting mechanism and outcome propagation to protocol
Protocol optimizations
Database research has been done on ways to get most of
the benefits of the two-phase commit protocol while reducing costs by protocol optimizations[1][2][3] and protocol operations saving under certain system’s behavior assumptions.
commitment, D2PC) protocol[2][6] is a variant of Tree
2PC with no predetermined coordinator. It subsumes
several optimizations that have been proposed earlier.
Agreement messages (Yes votes) start to propagate from
all the leaves, each leaf when completing its tasks on behalf of the transaction (becoming ready). An intermediate (non leaf) node sends when ready an agreement
message to the last (single) neighboring node from which
agreement message has not yet been received. The coordinator is determined dynamically by racing agreement
messages over the transaction tree, at the place where they
collide. They collide either at a transaction tree node, to
be the coordinator, or on a tree edge. In the latter case one
of the two edge’s nodes is elected as a coordinator (any
node). D2PC is time optimal (among all the instances of a
specific transaction tree, and any specific Tree 2PC protocol implementation; all instances have the same tree;
each instance has a different node as coordinator): By
choosing an optimal coordinator D2PC commits both the
coordinator and each cohort in minimum possible time,
allowing the earliest possible release of locked resources
in each transaction participant (tree node).
Presume abort and Presume commit Presumed abort or Presumed commit are common such
optimizations.[2][3][5] An assumption about the outcome
of transactions, either commit, or abort, can save both
messages and logging operations by the participants
during the 2PC protocol’s execution. For example, when
presumed abort, if during system recovery from failure
no logged evidence for commit of some transaction is
found by the recovery procedure, then it assumes that
the transaction has been aborted, and acts accordingly. 7.3.5 See also
This means that it does not matter if aborts are logged at
all, and such logging can be saved under this assumption.
• Atomic commit
Typically a penalty of additional operations is paid
• Commit (data management)
during recovery from failure, depending on optimization
• Three-phase commit protocol
• XA
• Paxos algorithm
• Two Generals’ Problem
[1] Philip A. Bernstein, Vassos Hadzilacos, Nathan Goodman
(1987): Concurrency Control and Recovery in Database
Systems, Chapter 7, Addison Wesley Publishing Company, ISBN 0-201-10715-5
[2] Gerhard Weikum,
Gottfried Vossen (2001):
Transactional Information Systems, Chapter 19, Elsevier, ISBN 1-55860-508-8
automata and proved that it is not resilient to a random
single site failure. The basic observation is that in 2PC,
while one site is in the “prepared to commit” state, the
other may be in either the “commit” or the “abort” state.
From this analysis, they developed 3PC to avoid such
states and it is thus resilient to such failures.
7.4.1 Protocol Description
In describing the protocol, we use terminology similar to
that used in the two-phase commit protocol. Thus we
have a single coordinator site leading the transaction and
a set of one or more cohorts being directed by the coordinator.
[3] Philip A. Bernstein, Eric Newcomer (2009): Principles of
Transaction Processing, 2nd Edition, Chapter 8, Morgan
Kaufmann (Elsevier), ISBN 978-1-55860-623-4
[4] C. Mohan, Bruce Lindsay and R. Obermarck (1986):
“Transaction management in the R* distributed database
management system”,ACM Transactions on Database Systems (TODS), Volume 11 Issue 4, Dec. 1986, Pages 378 396
[5] C. Mohan, Bruce Lindsay (1985): “Efficient commit
protocols for the tree of processes model of distributed
transactions”,ACM SIGOPS Operating Systems Review,
19(2),pp. 40-52 (April 1985)
[6] Yoav Raz (1995): “The Dynamic Two Phase Commitment (D2PC) protocol ",Database Theory — ICDT '95,
Lecture Notes in Computer Science, Volume 893/1995, pp.
162-176, Springer, ISBN 978-3-540-58907-5
External links
• Two Phase Commit protocol explained in Pictures
by exploreDatabase
7.4 Three-phase commit protocol
In computer networking and databases, the three-phase
commit protocol (3PC)[1] is a distributed algorithm
which lets all nodes in a distributed system agree to
commit a transaction. Unlike the two-phase commit protocol (2PC) however, 3PC is non-blocking. Specifically,
3PC places an upper bound on the amount of time required before a transaction either commits or aborts. This
property ensures that if a given transaction is attempting
to commit via 3PC and holds some resource locks, it will
release the locks after the timeout.
3PC was originally described by Dale Skeen and Michael
Stonebraker in their paper, “A Formal Model of Crash
Recovery in a Distributed System”.[1] In that work, they
modeled 2PC as a system of non-deterministic finite state
1. The coordinator receives a transaction request. If
there is a failure at this point, the coordinator aborts
the transaction (i.e. upon recovery, it will consider
the transaction aborted). Otherwise, the coordinator
sends a canCommit? message to the cohorts and
moves to the waiting state.
2. If there is a failure, timeout, or if the coordinator
receives a No message in the waiting state, the coordinator aborts the transaction and sends an abort
message to all cohorts. Otherwise the coordinator
will receive Yes messages from all cohorts within
the time window, so it sends preCommit messages
to all cohorts and moves to the prepared state.
3. If the coordinator succeeds in the prepared state, it
will move to the commit state. However if the coordinator times out while waiting for an acknowledgement from a cohort, it will abort the transaction. In
the case where all acknowledgements are received, Keidar and Dolev’s E3PC[2] algorithm eliminates this disthe coordinator moves to the commit state as well. advantage.
The protocol requires at least 3 round trips to complete,
needing a minimum of 3 round trip times (RTTs). This
is potentially a long latency to complete each transaction.
1. The cohort receives a canCommit? message from
the coordinator. If the cohort agrees it sends a Yes
message to the coordinator and moves to the pre- 7.4.4 References
pared state. Otherwise it sends a No message and
[1] Skeen, Dale; Stonebraker, M. (May 1983). “A Formal
aborts. If there is a failure, it moves to the abort
Model of Crash Recovery in a Distributed System”. IEEE
Transactions on Software Engineering 9 (3): 219–228.
2. In the prepared state, if the cohort receives an abort
message from the coordinator, fails, or times out
waiting for a commit, it aborts. If the cohort receives
a preCommit message, it sends an ACK message
back and awaits a final commit or abort.
[2] Keidar, Idit; Danny Dolev (December 1998). “Increasing
the Resilience of Distributed and Replicated Database
Systems”. Journal of Computer and System Sciences
(JCSS) 57 (3): 309–324. doi:10.1006/jcss.1998.1566.
3. If, after a cohort member receives a preCommit
message, the coordinator fails or times out, the co- 7.4.5 See also
hort member goes forward with the commit.
• Two-phase commit protocol
A Two-phase commit protocol cannot dependably recover from a failure of both the coordinator and a cohort
member during the Commit phase. If only the coordinator had failed, and no cohort members had received a
commit message, it could safely be inferred that no commit had happened. If, however, both the coordinator
and a cohort member failed, it is possible that the failed
cohort member was the first to be notified, and had actually done the commit. Even if a new coordinator is
selected, it cannot confidently proceed with the operation
until it has received an agreement from all cohort members ... and hence must block until all cohort members
The Three-phase commit protocol eliminates this problem by introducing the Prepared to commit state. If the
coordinator fails before sending preCommit messages,
the cohort will unanimously agree that the operation was
aborted. The coordinator will not send out a doCommit message until all cohort members have ACKed that
they are Prepared to commit. This eliminates the possibility that any cohort member actually completed the
transaction before all cohort members were aware of the
decision to do so (an ambiguity that necessitated indefinite blocking in the Two-phase commit protocol).
The main disadvantage to this algorithm is that it cannot recover in the event the network is segmented in any
manner. The original 3PC algorithm assumes a fail-stop
model, where processes fail by crashing and crashes can
be accurately detected, and does not work with network
partitions or asynchronous communication.
Chapter 8
8.1 Scalability
8.1.1 Measures
Scalability can be measured in various dimensions, such
Scalability is the capability of a system, network, or pro- as:
cess to handle a growing amount of work, or its potential
to be enlarged in order to accommodate that growth.[1]
• Administrative scalability: The ability for an increasFor example, it can refer to the capability of a system
ing number of organizations or users to easily share
to increase its total output under an increased load when
a single distributed system.
resources (typically hardware) are added. An analo• Functional scalability: The ability to enhance the
gous meaning is implied when the word is used in an
system by adding new functionality at minimal efeconomic context, where scalability of a company implies
that the underlying business model offers the potential for
economic growth within the company.
• Geographic scalability: The ability to maintain perScalability, as a property of systems, is generally difficult
formance, usefulness, or usability regardless of exto define[2] and in any particular case it is necessary to depansion from concentration in a local area to a more
fine the specific requirements for scalability on those didistributed geographic pattern.
mensions that are deemed important. It is a highly signif• Load scalability: The ability for a distributed system
icant issue in electronics systems, databases, routers, and
to easily expand and contract its resource pool to acnetworking. A system whose performance improves after
commodate heavier or lighter loads or number of inadding hardware, proportionally to the capacity added, is
puts. Alternatively, the ease with which a system or
said to be a scalable system.
component can be modified, added, or removed, to
An algorithm, design, networking protocol, program, or
accommodate changing load.
other system is said to scale if it is suitably efficient and
practical when applied to large situations (e.g. a large in• Generation scalability refers to the ability of a system
put data set, a large number of outputs or users, or a large
to scale up by using new generations of components.
number of participating nodes in the case of a distributed
Thereby, heterogeneous scalability is the ability to
system). If the design or system fails when a quantity inuse the components from different vendors.[4]
creases, it does not scale. In practice, if there are a large
number of things (n) that affect scaling, then resource requirements (for example, algorithmic time-complexity) 8.1.2 Examples
must grow less than n2 as n increases. An example is a
search engine, that must scale not only for the number of
• A routing protocol is considered scalable with reusers, but for the number of objects it indexes. Scalability
spect to network size, if the size of the necessary
refers to the ability of a site to increase in size as demand
routing table on each node grows as O(log N), where
N is the number of nodes in the network.
The concept of scalability is desirable in technology as
well as business settings. The base concept is consistent
– the ability for a business or technology to accept increased volume without impacting the contribution margin (= revenue − variable costs). For example, a given
piece of equipment may have a capacity for 1–1000 users,
while beyond 1000 users additional equipment is needed
or performance will decline (variable costs will increase
and reduce contribution margin).
• A scalable online transaction processing system or
database management system is one that can be upgraded to process more transactions by adding new
processors, devices and storage, and which can be
upgraded easily and transparently without shutting
it down.
• Some early peer-to-peer (P2P) implementations of
Gnutella had scaling issues. Each node query
flooded its requests to all peers. The demand on each
peer would increase in proportion to the total number of peers, quickly overrunning the peers’ limited
capacity. Other P2P systems like BitTorrent scale
well because the demand on each peer is independent of the total number of peers. There is no centralized bottleneck, so the system may expand indefinitely without the addition of supporting resources
(other than the peers themselves).
There are tradeoffs between the two models. Larger numbers of computers means increased management complexity, as well as a more complex programming model
and issues such as throughput and latency between nodes;
also, some applications do not lend themselves to a distributed computing model. In the past, the price difference between the two models has favored “scale up” computing for those applications that fit its paradigm, but recent advances in virtualization technology have blurred
that advantage, since deploying a new virtual system over
• The distributed nature of the Domain Name System a hypervisor (where possible) is almost always less expenallows it to work efficiently even when all hosts on sive than actually buying and installing a real one. Configthe worldwide Internet are served, so it is said to uring an existing idle system has always been less expen“scale well”.
sive than buying, installing, and configuring a new one,
regardless of the model.
Horizontal and vertical scaling
8.1.4 Database scalability
Methods of adding more resources for a particular application fall into two broad categories: horizontal and
A number of different approaches enable databases
vertical scaling.[5]
to grow to very large size while supporting an everincreasing rate of transactions per second. Not to be
• To scale horizontally (or scale out) means to add discounted, of course, is the rapid pace of hardware admore nodes to a system, such as adding a new com- vances in both the speed and capacity of mass storage deputer to a distributed software application. An vices, as well as similar advances in CPU and networking
example might involve scaling out from one Web speed.
server system to three. As computer prices have
One technique supported by most of the major database
dropped and performance continues to increase,
management system (DBMS) products is the partitioning
high-performance computing applications such as
of large tables, based on ranges of values in a key field.
seismic analysis and biotechnology workloads have
In this manner, the database can be scaled out across a
adopted low-cost "commodity" systems for tasks
cluster of separate database servers. Also, with the advent
that once would have required supercomputers. Sysof 64-bit microprocessors, multi-core CPUs, and large
tem architects may configure hundreds of small
SMP multiprocessors, DBMS vendors have been at the
computers in a cluster to obtain aggregate comforefront of supporting multi-threaded implementations
puting power that often exceeds that of computers
that substantially scale up transaction processing capacity.
based on a single traditional processor. The development of high-performance interconnects such as Network-attached storage (NAS) and Storage area netGigabit Ethernet, InfiniBand and Myrinet further fu- works (SANs) coupled with fast local area networks and
eled this model. Such growth has led to demand Fibre Channel technology enable still larger, more loosely
for software that allows efficient management and coupled configurations of databases and distributed commaintenance of multiple nodes, as well as hardware puting power. The widely supported X/Open XA stansuch as shared data storage with much higher I/O dard employs a global transaction monitor to coordinate
performance. Size scalability is the maximum num- distributed transactions among semi-autonomous XAber of processors that a system can accommodate.[4] compliant database resources. Oracle RAC uses a different model to achieve scalability, based on a “shared• To scale vertically (or scale up) means to add re- everything” architecture that relies upon high-speed consources to a single node in a system, typically in- nections between servers.
volving the addition of CPUs or memory to a single
computer. Such vertical scaling of existing systems
also enables them to use virtualization technology
more effectively, as it provides more resources for
the hosted set of operating system and application
modules to share. Taking advantage of such resources can also be called “scaling up”, such as expanding the number of Apache daemon processes
currently running. Application scalability refers to
the improved performance of running applications
on a scaled-up version of the system.[4]
While DBMS vendors debate the relative merits of their
favored designs, some companies and researchers question the inherent limitations of relational database management systems. GigaSpaces, for example, contends
that an entirely different model of distributed data access
and transaction processing, Space based architecture, is
required to achieve the highest performance and scalability. On the other hand, Base One makes the case for
extreme scalability without departing from mainstream
relational database technology.[6] For specialized applications, NoSQL architectures such as Google’s BigTable
can further enhance scalability. Google’s massively distributed Spanner technology, positioned as a successor
to BigTable, supports general-purpose database transactions and provides a more conventional SQL-based query
• while the storage cluster is partitioned, all parts remain responsive. There is a risk of conflicting updates.
8.1.6 Performance tuning versus hardware scalability
Strong versus eventual consistency
It is often advised to focus system design on hardware
In the context of scale-out data storage, scalability is defined as the maximum storage cluster size which guarantees full data consistency, meaning there is only ever one
valid version of stored data in the whole cluster, independently from the number of redundant physical data
copies. Clusters which provide “lazy” redundancy by
updating copies in an asynchronous fashion are called
'eventually consistent'. This type of scale-out design is
suitable when availability and responsiveness are rated
higher than consistency, which is true for many web file
hosting services or web caches (if you want the latest version, wait some seconds for it to propagate). For all classical transaction-oriented applications, this design should
be avoided.[8]
Many open source and even commercial scale-out storage clusters, especially those built on top of standard
PC hardware and networks, provide eventual consistency
only. Idem some NoSQL databases like CouchDB and
others mentioned above. Write operations invalidate
other copies, but often don't wait for their acknowledgements. Read operations typically don't check every redundant copy prior to answering, potentially missing the
preceding write operation. The large amount of metadata signal traffic would require specialized hardware
and short distances to be handled with acceptable performance (i.e. act like a non-clustered storage device or
scalability rather than on capacity. It is typically cheaper
to add a new node to a system in order to achieve improved performance than to partake in performance tuning to improve the capacity that each node can handle.
But this approach can have diminishing returns (as discussed in performance engineering). For example: suppose 70% of a program can be sped up if parallelized and
run on multiple CPUs instead of one. If α is the fraction
of a calculation that is sequential, and 1−α is the fraction
that can be parallelized, the maximum speedup that can
be achieved by using P processors is given according to
Amdahl’s Law:
α+ 1−α
Substituting the value for this example, using 4 processors
we get
0.3+ 1−0.3
= 2.105 .
If we double the compute power to 8 processors we get
0.3+ 1−0.3
= 2.581 .
Doubling the processing power has only improved the
speedup by roughly one-fifth. If the whole problem was
parallelizable, we would, of course, expect the speed up
to double also. Therefore, throwing in more hardware is
not necessarily the optimal approach.
8.1.7 Weak versus strong scaling
Whenever strong data consistency is expected, look for
In the context of high performance computing there are
these indicators:
two common notions of scalability:
• the use of InfiniBand, Fibrechannel or similar lowlatency networks to avoid performance degradation
with increasing cluster size and number of redundant
• The first is strong scaling, which is defined as how the
solution time varies with the number of processors
for a fixed total problem size.
• short cable lengths and limited physical extent,
avoiding signal runtime performance degradation.
• The second is weak scaling, which is defined as how
the solution time varies with the number of processors for a fixed problem size per processor.[9]
• majority / quorum mechanisms to guarantee data
consistency whenever parts of the cluster become in8.1.8
Indicators for eventually consistent designs (not suitable
for transactional applications!) are:
• write performance increases linearly with the number of connected devices in the cluster.
See also
• Asymptotic complexity
• Computational complexity theory
• Data Defined Storage
• Extensibility
• Gustafson’s law
• List of system quality attributes
• Load balancing (computing)
• Lock (computer science)
• Parallel computing
• Scalable Video Coding (SVC)
• Similitude (model)
[1] Bondi, André B. (2000). Characteristics of scalability and
their impact on performance. Proceedings of the second
international workshop on Software and performance WOSP '00. p. 195. doi:10.1145/350391.350432. ISBN
[2] See for instance, Hill, Mark D. (1990). “What is scalability?". ACM SIGARCH Computer Architecture News
18 (4): 18. doi:10.1145/121973.121975. and Duboc,
Leticia; Rosenblum, David S.; Wicks, Tony (2006). A
framework for modelling and analysis of software systems scalability. Proceeding of the 28th international conference on Software engineering - ICSE '06. p. 949.
doi:10.1145/1134285.1134460. ISBN 1595933751.
[3] Laudon, Kenneth Craig; Traver, Carol Guercio (2008). Ecommerce: Business, Technology, Society. Pearson Prentice Hall/Pearson Education. ISBN 9780136006459.
[4] Hesham El-Rewini and Mostafa Abd-El-Barr (Apr 2005).
Advanced Computer Architecture and Parallel Processing.
John Wiley & Son. p. 66. ISBN 978-0-471-47839-3.
Retrieved Oct 2013.
8.1.10 External links
• Architecture of a Highly Scalable NIO-Based Server
- an article about writing scalable server in Java
• Links to diverse learning resources - page curated by
the memcached project.
• Scalable Definition - by The Linux Information
Project (LINFO)
• Scale in Distributed Systems B. Clifford Neumann,
In: Readings in Distributed Computing Systems,
IEEE Computer Society Press, 1994
8.2 Shard (database architecture)
A database shard is a horizontal partition of data in a
database or search engine. Each individual partition is
referred to as a shard or database shard. Each shard
is held on a separate database server instance, to spread
Some data within a database remains present in all
shards,[notes 1] but some only appears in a single shard.
Each shard (or server) acts as the single source for this
subset of data.[1]
8.2.1 Database architecture
Horizontal partitioning is a database design principle
whereby rows of a database table are held separately,
rather than being split into columns (which is what
normalization and vertical partitioning do, to differing
extents). Each partition forms part of a shard, which
may in turn be located on a separate database server or
physical location.
There are numerous advantages to the horizontal partitioning approach. Since the tables are divided and distributed into multiple servers, the total number of rows
in each table in each database is reduced. This reduces
index size, which generally improves search performance.
A database shard can be placed on separate hardware, and
multiple shards can be placed on multiple machines. This
[6] Base One (2007). “Database Scalability - Dispelling
enables a distribution of the database over a large number
myths about the limits of database-centric architecture”.
of machines, which means that the load can be spread out
Retrieved May 23, 2007.
over multiple machines, greatly improving performance.
[7] “Spanner: Google’s Globally-Distributed Database” In addition, if the database shard is based on some real(PDF). OSDI'12 Proceedings of the 10th USENIX con- world segmentation of the data (e.g., European customers
ference on Operating Systems Design and Implementa- v. American customers) then it may be possible to infer
tion. 2012. pp. 251–264. ISBN 978-1-931971-96-6. the appropriate shard membership easily and automatiRetrieved September 30, 2012.
cally, and query only the relevant shard.[2] Disadvantages
include :
[5] Michael, Maged; Moreira, Jose E.; Shiloach, Doron; Wisniewski, Robert W. (March 26, 2007). Scale-up x Scaleout: A Case Study using Nutch/Lucene. 2007 IEEE International Parallel and Distributed Processing Symposium.
p. 1. doi:10.1109/IPDPS.2007.370631. ISBN 1-42440909-8.
[8] “Eventual consistency by Werner Vogels”.
[9] “CSE - CSE - The Weak Scaling of DL_POLY 3”. Retrieved 2014-03-08.
• A heavier reliance on the interconnect between
• Increased latency when querying, especially where This makes replication across multiple servers easy (simmore than one shard must be searched.
ple horizontal partitioning does not). It is also useful for
worldwide distribution of applications, where communi• Data or indexes are often only sharded one
cations links between data centers would otherwise be a
way, so that some searches are optimal, and
others are slow or impossible.
There is also a requirement for some notification and
• Issues of consistency and durability due to the more replication mechanism between schema instances, so that
complex failure modes of a set of servers, which the unpartitioned tables remain as closely synchronized as
often result in systems making no guarantees about the application demands. This is a complex choice in the
cross-shard consistency or durability.
architecture of sharded systems: approaches range from
making these effectively read-only (updates are rare and
In practice, sharding is complex. Although it has been
batched), to dynamically replicated tables (at the cost of
done for a long time by hand-coding (especially where
reducing some of the distribution benefits of sharding)
rows have an obvious grouping, as per the example
and many options in between.
above), this is often inflexible. There is a desire to support sharding automatically, both in terms of adding code
support for it, and for identifying candidates to be sharded 8.2.3 Support for shards
separately. Consistent hashing is one form of automatic
sharding to spread large loads across multiple smaller ser- Apache HBase HBase supports automatic sharding.[4]
vices and servers.[3]
Where distributed computing is used to separate load be- Azure SQL Database Elastic Database tools Elastic
tween multiple servers (either for performance or reliaDatabase tools enables the data-tier of an applibility reasons), a shard approach may also be useful.
cation to scale out and in via industry-standard
sharding practices[5]
Shards compared to horizontal parCUBRID CUBRID supports sharding from version 9.0
Horizontal partitioning splits one or more tables by row, dbShards CodeFutures[6]dbShards is a product dedicated
to database shards.
usually within a single instance of a schema and a database
server. It may offer an advantage by reducing index size
(and thus search effort) provided that there is some obvi- Elasticsearch Elasticsearch enterprise search server
provides sharding capabilities.[7]
ous, robust, implicit way to identify in which table a particular row will be found, without first needing to search
the index, e.g., the classic example of the 'CustomersEast' eXtreme Scale eXtreme Scale is a cross-process inand 'CustomersWest' tables, where their zip code already
memory key/value datastore (a variety of NoSQL
indicates where they will be found.
datastore). It uses sharding to achieve scalability
across processes for both data and MapReduce-style
Sharding goes beyond this: it partitions the problematic
parallel processing.[8]
table(s) in the same way, but it does this across potentially
multiple instances of the schema. The obvious advantage
would be that search load for the large partitioned table Hibernate ORM Hibernate Shards provides support for
can now be split across multiple servers (logical or physishards, although there has been little activity since
cal), not just multiple indexes on the same logical server.
Splitting shards across multiple isolated instances requires more than simple horizontal partitioning. The IBM Informix IBM has supported sharding in Informix
since version 12.1 xC1 as part of the MACH11
hoped-for gains in efficiency would be lost, if querying
technology. Informix 12.10 xC2 added full comthe database required both instances to be queried, just to
patibility with MongoDB drivers, allowing the mix
retrieve a simple dimension table. Beyond partitioning,
of regular relational tables with NoSQL collections,
sharding thus splits large partitionable tables across the
while still supporting sharding, failover and ACID
servers, while smaller tables are replicated as complete
This is also why sharding is related to a shared nothing
architecture—once sharded, each shard can live in a to- MonetDB the open-source column-store MonetDB supports read-only sharding as its July 2015 release.[13]
tally separate logical schema instance / physical database
server / data center / continent. There is no ongoing need
to retain shared access (from between shards) to the other MongoDB MongoDB supports sharding from version
unpartitioned tables in other shards.
MySQL Cluster Auto-Sharding: Database is automatically and transparently partitioned across low-cost
commodity nodes, allowing scale-out of read and
write queries, without requiring changes to the
8.2.4 Disadvantages of sharding
Sharding a database table before it has been optimized
locally causes premature complexity. Sharding should be
used only when all other options for optimization are inadequate. The introduced complexity of database shardMySQL Fabric (part of MySQL utilities) includes ing causes the following potential problems:
support for sharding.[15]
Oracle NoSQL Database
• Increased complexity of SQL - Increased bugs because the developers have to write more complicated
SQL to handle sharding logic.
Oracle NoSQL Database supports automatic sharding
and elastic, online expansion of the cluster (adding more
• Sharding introduces complexity - The sharding
software that partitions, balances, coordinates, and
ensures integrity can fail.
OrientDB OrientDB supports sharding from version 1.7
• Single point of failure - Corruption of one shard
due to network/hardware/systems problems causes
failure of the entire table.
pg_shard a sharding extension for PostgreSQL. It shards
and replicates PostgreSQL tables for horizontal
scale and for high availability. The extension also
seamlessly distributes SQL statements without requiring any changes to applications.[16]
Plugin for Grails Grails supports sharding using the
Grails Sharding Plugin.[17]
Ruby ActiveRecord Octopus works as a database
sharding and replication extension for the ActiveRecord ORM.
• Failover servers more complex - Failover servers
must themselves have copies of the fleets of database
• Backups more complex - Database backups of
the individual shards must be coordinated with the
backups of the other shards.
• Operational
Adding/removing indexes, adding/deleting columns,
modifying the schema becomes much more difficult.
ScaleBase’s Data Traffic Manager ScaleBase’s Data
Traffic Manager is a software product dedicated
to automating MySQL database sharding without These historical complications of do-it-yourself sharding
are now being addressed by independent software vendors
requiring changes to applications.[18]
who provide autosharding solutions.
Shard Query Open Source parallel query engine for
8.2.5 Etymology
Solr Search Server Solr enterprise search server pro- The word “shard” in a database context may have been
vides sharding capabilities.[20]
introduced by the CCA's “System for Highly Available
Replicated Data”.[23] There has been speculation[24] that
Spanner Spanner, Google’s global-scale distributed the term might be derived from the 1997 MMORPG
database, shards data across multiple Paxos state Ultima Online, but the SHARD database system predates
machines to scale to “millions of machines across this by at least nine years.
hundreds of datacenters and trillions of database
However, the SHARD system appears[25] to have used its
redundant hardware only for replication and not for horizontal partitioning. It is not known whether present-day
SQLAlchemy ORM SQLAlchemy is a data-mapper for use of the term “shard” is derived from the CCA system,
the Python programming language that provides but in any case it refers to a different use of redundant
sharding capabilities.[22]
hardware in database systems.
The DWH of Teradata was the first massive
parallel database.
8.2.6 See also
• Shared nothing architecture
[1] Typically 'supporting' data such as dimension tables
[1] Pramod J. Sadalage; Martin Fowler (2012), “4: Distribution Models”, NoSQL Distilled, ISBN 0321826620
[24] Koster, Raph (2009-01-08). “Database “sharding” came
from UO?". Raph Koster’s Website. Retrieved 2015-0117.
[25] Sarin & Lynch, Discarding Obsolete Information in a
Replicated Database System, IEEE Transactions on Software Engineering vol SE-13 no 1, January 1987
[2] Rahul Roy (July 28, 2008). “Shard - A Database Design”.
[3] Ries, Eric. “Sharding for Startups”.
[4] “Apache HBase Sharding”.
[5] “Introducing Elastic Scale preview for Azure SQL
[6] “dbShards product overview”.
8.2.8 External links
• Informix JSON data sharding
8.3 Optimistic concurrency control
[7] “Index Shard Allocation”.
[9] “Hibernate Shards”. 2007-02-08.
[10] “Hibernate Shards”.
[11] “New Grid queries for Informix”.
[12] “NoSQL support in Informix”.
[13] “MonetDB July2015 Released”. 31 August 2015.
[14] “MySQL Cluster Features & Benefits”. 2012-11-23.
[15] “MySQL Fabric sharding quick start guide”.
[16] “pg_shard PostgreSQL extension”.
[17] “Grails Sharding Plugin”.
[18] “ScaleBase’s Data Traffic Manager product architecture
[19] “Shard Query”.
[20] “Distributed Search”.
[21] Corbett, James C; Dean, Jeffrey; Epstein, Michael; Fikes,
Andrew; Frost, Christopher; Furman, JJ; Ghemawat, Sanjay; Heiser, Christopher; Hochschild, Peter; Hsieh, Wilson; Kanthak, Sebastian; Kogan, Eugene; Li, Hongyi;
Lloyd, Alexander; Melnik, Sergey; Mwaura, David; Nagle, David; Quinlan, Sean; Rao, Rajesh; Rolig, Lindsay; Saito, Yasushi; Szymaniak, Michal; Taylor, Christopher; Wang, Ruth; Woodford, Dale. “Spanner: Google’s
Globally-Distributed Database” (PDF). Proceedings of
OSDI 2012. Google. Retrieved 24 February 2014. |first8=
missing |last8= in Authors list (help)
[22] “Basic example of using the SQLAlchemy Sharding
[23] Sarin, DeWitt & Rosenburg, Overview of SHARD: A
System for Highly Available Replicated Data, Technical
Report CCA-88-01, Computer Corporation of America,
May 1988
Optimistic concurrency control (OCC) is a
concurrency control method applied to transactional
systems such as relational database management systems
and software transactional memory. OCC assumes that
multiple transactions can frequently complete without
interfering with each other. While running, transactions
use data resources without acquiring locks on those
resources. Before committing, each transaction verifies
that no other transaction has modified the data it has
read. If the check reveals conflicting modifications, the
committing transaction rolls back and can be restarted.[1]
Optimistic concurrency control was first proposed by
H.T. Kung.[2]
OCC is generally used in environments with low data contention. When conflicts are rare, transactions can complete without the expense of managing locks and without having transactions wait for other transactions’ locks
to clear, leading to higher throughput than other concurrency control methods. However, if contention for data
resources is frequent, the cost of repeatedly restarting
transactions hurts performance significantly; it is commonly thought that other concurrency control methods
have better performance under these conditions. However, locking-based (“pessimistic”) methods also can deliver poor performance because locking can drastically
limit effective concurrency even when deadlocks are
8.3.1 OCC phases
More specifically, OCC transactions involve these phases:
• Begin: Record a timestamp marking the transaction’s beginning.
• Modify: Read database values, and tentatively write
• Validate: Check whether other transactions have
modified data that this transaction has used (read
or written). This includes transactions that completed after this transaction’s start time, and optionally, transactions that are still active at validation
• Commit/Rollback: If there is no conflict, make all
changes take effect. If there is a conflict, resolve it,
typically by aborting the transaction, although other
resolution schemes are possible. Care must be taken
to avoid a TOCTTOU bug, particularly if this phase
and the previous one are not performed as a single
atomic operation.
Web usage
• Mimer SQL is a DBMS that only implements optimistic concurrency control.[10]
• Google App Engine data store uses OCC.[11]
• The Elasticsearch search engine supports OCC via
the version attribute.[12]
• The MonetDB column-oriented database management system's transaction management scheme is
based on OCC.[13]
• Most implementations of software transactional
memory use optimistic locking.
• Redis provides
The stateless nature of HTTP makes locking infeasible
for web user interfaces. It’s common for a user to start 8.3.3 See also
editing a record, then leave without following a “cancel”
or “logout” link. If locking is used, other users who at• Server Message Block#Opportunistic locking
tempt to edit the same record must wait until the first
user’s lock times out.
HTTP does provide a form of built-in OCC: The GET
method returns an ETag for a resource and subsequent
PUTs use the ETag value in the If-Match headers; while
the first PUT will succeed, the second will not, as the
value in If-Match is based on the first version of the
Some database management systems offer OCC natively
- without requiring special application code. For others, the application can implement an OCC layer outside
of the database, and avoid waiting or silently overwriting records. In such cases, the form includes a hidden
field with the record’s original content, a timestamp, a sequence number, or an opaque token. On submit, this is
compared against the database. If it differs, the conflict
resolution algorithm is invoked.
8.3.4 References
[1] Johnson, Rohit (2003). “Common Data Access Issues”.
Expert One-on-One J2EE Design and Development. Wrox
Press. ISBN 0-7645-4385-7.
[2] Kung, H.T. (1981). “On Optimistic Methods for Concurrency Control”. ACM Transactions on Database Systems.
[3] “Editing the Web - Detecting the Lost Update Problem
Using Unreserved Checkout”. W3C Note. 10 May 1999.
[4] Help:Edit conflict
[5] “Bugzilla: FAQ: Administrative Questions”. MozillaWiki.
11 April 2012.
[6] “Module ActiveRecord::Locking”.
Rails Framework
[7] “Object Relational Mapping (GORM)". Grails Framework Documentation.
• MediaWiki's edit pages use OCC.[4]
[8] “Transaction Processing”.
UNIX Edition.
GT.M Programmers Guide
• Bugzilla uses OCC; edit conflicts are called “mid-air
[9] “Tip 19 – How to use Optimistic Concurrency with the
Entity Framework”. MSDN Blogs. 19 May 2009.
• The Ruby on Rails framework has an API for
• Most revision control systems support the “merge”
model for concurrency, which is OCC.
• The Grails framework uses OCC in its default [10] “Transaction Concurrency - Optimistic Concurrency Control”. Mimer Developers - Features. 26 February 2010.
[11] “The Datastore”. What Is Google App Engine?. 27 August
• The GT.M database engine uses OCC for manag2010.
ing transactions[8] (even single updates are treated
[12] “Elasticsearch - Guide - Index API”. Elasticsearch Guide.
as mini-transactions).
22 March 2012.
• Microsoft's Entity Framework (including CodeFirst) has built-in support for OCC based on a binary
timestamp value.[9]
[13] “Transactions - MonetDB”. 16 January 2013.
[14] “Transactions in Redis”.
External links
Hash partitioning The value of a hash function determines membership in a partition. Assuming there
are four partitions, the hash function could return a
• Kung, H. T.; John T. Robinson (June 1981). “On
value from 0 to 3.
optimistic methods for concurrency control”. ACM
Transactions on Database Systems 6 (2): 213–226.
Composite partitioning allows for certain combinadoi:10.1145/319566.319567.
tions of the above partitioning schemes, by for ex• Enterprise JavaBeans, 3.0, By Bill Burke, Richard
ample first applying a range partitioning and then
Monson-Haefel, Chapter 16. Transactions, Section
a hash partitioning. Consistent hashing could be
16.3.5. Optimistic Locking, Publisher: O'Reilly,
considered a composite of hash and list partitioning
Pub Date: May 16, 2006,Print ISBN 0-596-00978where the hash reduces the key space to a size that
can be listed.
• Hollmann, Andreas (May 2009). “Multi-Isolation:
Virtues and Limitations” (PDF). Multi-Isolation 8.4.3 Partitioning methods
(what is between pessimistic and optimistic locking).
01069 Gutzkovstr. 30/F301.2, Dresden: Happy- The partitioning can be done by either building separate
Guys Software GbR. p. 8. Retrieved 2013-05-16.
smaller databases (each with its own tables, indices, and
transaction logs), or by splitting selected elements, for example just one table.
8.4 Partition (database)
Horizontal partitioning (also see shard) involves
putting different rows into different tables. Perhaps cusA partition is a division of a logical database or tomers with ZIP codes less than 50000 are stored in Cusits constituent elements into distinct independent parts. tomersEast, while customers with ZIP codes greater than
Database partitioning is normally done for manageabil- or equal to 50000 are stored in CustomersWest. The two
partition tables are then CustomersEast and Customerity, performance or availability reasons.
sWest, while a view with a union might be created over
both of them to provide a complete view of all customers.
Benefits of multiple partitions
A popular and favourable application of partitioning is in
a distributed database management system. Each partition may be spread over multiple nodes, and users at the
node can perform local transactions on the partition. This
increases performance for sites that have regular transactions involving certain views of data, whilst maintaining
availability and security.
Partitioning criteria
Current high end relational database management systems provide for different criteria to split the database.
They take a partitioning key and assign a partition based
on certain criteria. Common criteria are:
Vertical partitioning involves creating tables with fewer
columns and using additional tables to store the remaining
columns.[1] Normalization also involves this splitting of
columns across tables, but vertical partitioning goes beyond that and partitions columns even when already normalized. Different physical storage might be used to realize vertical partitioning as well; storing infrequently used
or very wide columns on a different device, for example,
is a method of vertical partitioning. Done explicitly or
implicitly, this type of partitioning is called “row splitting” (the row is split by its columns). A common form
of vertical partitioning is to split dynamic data (slow to
find) from static data (fast to find) in a table where the
dynamic data is not used as often as the static. Creating
a view across the two newly created tables restores the
original table with a performance penalty, however performance will increase when accessing the static data e.g.
for statistical analysis.
Range partitioning Selects a partition by determining
if the partitioning key is inside a certain range. An
example could be a partition for all rows where 8.4.4 See also
the column zipcode has a value between 70000 and
• Shard (database architecture)
List partitioning A partition is assigned a list of values.
If the partitioning key has one of these values, the 8.4.5 References
partition is chosen. For example all rows where the
column Country is either Iceland, Norway, Sweden, [1] Vertical Partitioning Algorithms for Database Design, by
Finland or Denmark could build a partition for the
Shamkant Navathe, Stefano Ceri, Gio Wiederhold, and
Nordic countries.
Jinglie Dou, Stanford University 1984
External links
• IBM DB2 partitioning
• MySQL partitioning
• Oracle partitioning
• SQL Server partitions
• PostgreSQL partitioning
• Sybase ASE 15.0 partitioning
• MongoDB partitioning
• ScimoreDB partitioning
• VoltDB partitioning
booking a flight, a rental car and a hotel. Since booking the flight might take up to a day to get a confirmation,
two-phase commit is not applicable here, it will lock the
resources for this long. In this case more sophisticated
techniques that involve multiple undo levels are used. The
way you can undo the hotel booking by calling a desk and
cancelling the reservation, a system can be designed to
undo certain operations (unless they are irreversibly finished).
In practice, long-lived distributed transactions are implemented in systems based on Web Services. Usually these
transactions utilize principles of Compensating transactions, Optimism and Isolation Without Locking. X/Open
standard does not cover long-lived DTP.
Several modern technologies, including Enterprise Java
Beans (EJBs) and Microsoft Transaction Server (MTS)
fully support distributed transaction standards.
8.5 Distributed transaction
8.5.1 See also
A distributed transaction is a database transaction in
which two or more network hosts are involved. Usually, Java Transaction API (JTA)
hosts provide transactional resources, while the transaction manager is responsible for creating and manag8.5.2 References
ing a global transaction that encompasses all operations
against such resources. Distributed transactions, as any
• “Web-Services Transactions”. Web-Services Transother transactions, must have all four ACID (atomicactions. Retrieved May 2, 2005.
ity, consistency, isolation, durability) properties, where
atomicity guarantees all-or-nothing outcomes for the unit
• “Nuts And Bolts Of Transaction Processing”. Artiof work (operations bundle).
cle about Transaction Management. Retrieved May
3, 2005.
Open Group, a vendor consortium, proposed the
X/Open Distributed Transaction Processing (DTP)
Model (X/Open XA), which became a de facto standard
for behavior of transaction model components.
• “A Detailed Comparison of Enterprise JavaBeans
(EJB) & The Microsoft Transaction Server (MTS)
Databases are common transactional resources and, often, transactions span a couple of such databases. In this
case, a distributed transaction can be seen as a database 8.5.3 Further reading
transaction that must be synchronized (or provide ACID
• Gerhard Weikum, Gottfried Vossen, Transactional
properties) among multiple participating databases which
information systems: theory, algorithms, and the
are distributed among different physical locations. The
practice of concurrency control and recovery, Morisolation property (the I of ACID) poses a special chalgan Kaufmann, 2002, ISBN 1-55860-508-8
lenge for multi database transactions, since the (global)
serializability property could be violated, even if each
database provides it (see also global serializability). In
practice most commercial database systems use strong
strict two phase locking (SS2PL) for concurrency control,
which ensures global serializability, if all the participating databases employ it. (see also commitment ordering
for multidatabases.)
A common algorithm for ensuring correct completion of
a distributed transaction is the two-phase commit (2PC).
This algorithm is usually applied for updates able to
commit in a short period of time, ranging from couple
of milliseconds to couple of minutes.
There are also long-lived distributed transactions, for example a transaction to book a trip, which consists of
Chapter 9
9.1 Redis
• Hash tables where keys and values are strings
This article is about Redis software. For Redis people,
see Romani people.
• HyperLogLogs used for approximated set cardinality size estimation.
Redis is a data structure server. It is open-source,
networked, in-memory, and stores keys with optional
durability. The development of Redis has been sponsored
by Redis Labs since June 2015.[3] Before that, it was
sponsored by Pivotal Software[4] and by VMware.[5][6]
According to the monthly ranking by,
Redis is the most popular key-value database.[7] Redis
has also been ranked the #1 NoSQL (and #4 database)
in User Satisfaction and Market Presence based on
user reviews,[8] the most popular NoSQL database in
containers,[9] and the #2 NoSQL among Top 50 Developer Tools & Services.[10] The name Redis means REmote DIctionary Server.[11]
Supported languages
Many languages have Redis bindings, including:[12]
ActionScript, C, C++, C#, Chicken Scheme, Clojure,
Common Lisp, D, Dart, Erlang, Go, Haskell, Haxe, Io,
Java, JavaScript (Node.js), Julia, Lua, Objective-C, Perl,
PHP, Pure Data, Python, R,[13] Racket, Ruby, Rust,
Scala, Smalltalk and Tcl.
Data types
The type of a value determines what operations (called
commands) are available for the value itself. Redis supports high-level, atomic, server-side operations like intersection, union, and difference between sets and sorting of
lists, sets and sorted sets.
9.1.3 Persistence
Redis typically holds the whole dataset in memory. Versions up to 2.4 could be configured to use what they refer
to as virtual memory[14] in which some of the dataset is
stored on disk, but this feature is deprecated. Persistence
is now reached in two different ways: one is called snapshotting, and is a semi-persistent durability mode where
the dataset is asynchronously transferred from memory
to disk from time to time, written in RDB dump format.
Since version 1.1 the safer alternative is AOF, an appendonly file (a journal) that is written as operations modifying the dataset in memory are processed. Redis is able to
rewrite the append-only file in the background in order to
avoid an indefinite growth of the journal.
By default, Redis syncs data to the disk at least every
2 seconds, with more or less robust options available if
needed. In the case of a complete system failure on default settings, only a few seconds of data would be lost.[15]
Redis maps keys to types of values. A key difference
between Redis and other structured storage systems is
that Redis supports not only strings, but also abstract data 9.1.4 Replication
Redis supports master-slave replication. Data from any
Redis server can replicate to any number of slaves. A
• Lists of strings
slave may be a master to another slave. This allows Re• Sets of strings (collections of non-repeating un- dis to implement a single-rooted replication tree. Redis slaves can be configured to accept writes, permitting
sorted elements)
intentional and unintentional inconsistency between in• Sorted sets of strings (collections of non-repeating stances. The Publish/Subscribe feature is fully impleelements ordered by a floating-point number called mented, so a client of a slave may SUBSCRIBE to a chanscore)
nel and receive a full feed of messages PUBLISHed to
9.1. REDIS
the master, anywhere up the replication tree. Replica- [10] Top 50 Developer Tools and Services of 2014
tion is useful for read (but not write) scalability or data
[11] “FAQ, Redis”.
[12] Redis language bindings
[13] CRAN – Package rredis
When the durability of data is not needed, the in-memory
nature of Redis allows it to perform extremely well com- [14] Redis documentation “Virtual Memory”,, accessed January 18, 2011.
pared to database systems that write every change to disk
before considering a transaction committed.[11] There is
no notable speed difference between write and read op- [15] Redis persistence demystified, 26 March 2012, antirez
erations. Redis operates as a single process and singlethreaded. Therefore a single Redis instance cannot utilize
[16] ReplicationHowto – redis – A persistent key-value
parallel execution of tasks e.g. stored procedures (Lua
database with built-in net interface written in ANSI-C for
Posix systems – Google Project Hosting
The Redis project has a cluster specification,[17] but the
cluster feature is currently in Beta stage.[18] According to
a news post by Redis creator Sanfilippo, the first production version of Redis cluster (planned for beta release at
end of 2013),[19] will support automatic partitioning of
the key space and hot resharding, but will support only
single key operations.[20] In the future Redis Cluster is
planned to support up to 1000 nodes, fault tolerance with
heartbeat and failure detection, incremental versioning
(“epochs”) to prevent conflicts, slave election and promotion to master, and publish/subscribe between all cluster
See also
[1] An interview with Salvatore Sanfilippo, creator of Redis, working out of Sicily, January 4, 2011, by Stefano
Bernardi, EU-Startups
[2] Pivotal People—Salvatore Sanfilippo, Inventor of Redis,
July 17, 2013, By Stacey Schneider, Pivotal P.O.V.
[17] Redis Cluster Specification,, Retrieved 2013-1225.
[18] Redis Cluster Tutorial,, Retrieved 2014-06-14.
[19] Redis Download Page,, Retrieved 2013-12-25.
[20] News about Redis: 2.8 is shaping, I'm back on Cluster,
Antirez Weblog - Salvatore Sanfilippo, Retrieved 201312-25.
[21] Redis Cluster - a Pragmatic Approach to Distribution,, Retrieved 2013-12-25.
• Jeremy Zawodny, Redis: Lightweight key/value Store
That Goes the Extra Mile, Linux Magazine, August
31, 2009
• Isabel Drost and Jan Lehnard (29 October 2009),
Happenings: NoSQL Conference, Berlin, The H.
Slides for the Redis presentation. Summary.
• Billy Newport (IBM): "Evolving the Key/Value Programming Model to a Higher Level" Qcon Conference 2009 San Francisco.
• A Mishra: "Install and configure Redis on Centos/
Fedora server".
[4] Redis Sponsors – Redis
[5] VMware: the new Redis home
9.1.9 External links
[6] VMWare: The Console: VMware hires key developer for
• Official website
[7] DB-Engines Ranking of Key-value Stores
• redis on GitHub
[8] Best NoSQL Databases: Fall 2015 Report from G2
• Redis Mailing List Archives
[9] The Current State of Container Usage
• Redis Articles Collection
9.2 MongoDB
MongoDB (from humongous) is a cross-platform
document-oriented database. Classified as a NoSQL
database, MongoDB eschews the traditional table-based
relational database structure in favor of JSON-like documents with dynamic schemas (MongoDB calls the format
BSON), making the integration of data in certain types of
applications easier and faster. Released under a combination of the GNU Affero General Public License and
the Apache License, MongoDB is free and open-source
First developed by the software company MongoDB Inc.
in October 2007 as a component of a planned platform
as a service product, the company shifted to an open
source development model in 2009, with MongoDB offering commercial support and other services.[2] Since
then, MongoDB has been adopted as backend software
by a number of major websites and services, including
Craigslist, eBay, and Foursquare among others. As of
July 2015, MongoDB is the fourth most popular type of
database management system, and the most popular for
document stores.[3]
MongoDB provides high availability with replica sets.[16]
A replica set consists of two or more copies of the data.
Each replica set member may act in the role of primary or
secondary replica at any time. The primary replica performs all writes and reads by default. Secondary replicas
maintain a copy of the data of the primary using builtin replication. When a primary replica fails, the replica
set automatically conducts an election process to determine which secondary should become the primary. Secondaries can also perform read operations, but the data is
eventually consistent by default.
Load balancing
MongoDB scales horizontally using sharding.[17] The user
chooses a shard key, which determines how the data in a
collection will be distributed. The data is split into ranges
(based on the shard key) and distributed across multiple
shards. (A shard is a master with one or more slaves.)
MongoDB can run over multiple servers, balancing the
load and/or duplicating data to keep the system up and
running in case of hardware failure. Automatic configuration is easy to deploy, and new machines can be added
to a running database.
File storage
Main features
MongoDB can be used as a file system, taking advantage
of load balancing and data replication features over multiple machines for storing files.
Some of the features include:[14]
This function, called Grid File System,[18] is included
with MongoDB drivers and available for development
languages (see "Language Support" for a list of supported
languages). MongoDB exposes functions for file manipulation and content to developers. GridFS is used, for example, in plugins for NGINX[19] and lighttpd.[20] Instead
of storing a file in a single document, GridFS divides a
file into parts, or chunks, and stores each of those chunks
as a separate document.[21]
Instead of taking a business subject and breaking it up
into multiple relational structures, MongoDB can store
the business subject in the minimal number of documents. For example, instead of storing title and author
information in two distinct relational structures, title, author, and other title-related information can all be stored
in a single document called Book.[15]
In a multi-machine MongoDB system, files can be distributed and copied multiple times between machines
transparently, thus effectively creating a load-balanced
Ad hoc queries
and fault-tolerant system.
MongoDB supports search by field, range queries, regular
expression searches. Queries can return specific fields of
documents and also include user-defined JavaScript funcMapReduce can be used for batch processing of data and
aggregation operations. The aggregation framework enables users to obtain the kind of results for which the SQL
GROUP BY clause is used.
Any field in a MongoDB document can be indexed (in- Server-side JavaScript execution
dices in MongoDB are conceptually similar to those in
RDBMSes). Secondary indices are also available.
JavaScript can be used in queries, aggregation functions
(such as MapReduce), and sent directly to the database
to be executed.
Capped collections
MongoDB supports fixed-size collections called capped
collections. This type of collection maintains insertion
order and, once the specified size has been reached, behaves like a circular queue.
In some failure scenarios where an application can access
two distinct MongoDB processes, but these processes
cannot access each other, it is possible for MongoDB
to return stale reads. In this scenario it is also possible
for MongoDB to acknowledge writes that will be rolled
Record insertion in MongoDB with Robomongo 0.8.5.
Management and graphical front-ends
Most administration is done from command line tools
such as the mongo shell because MongoDB does not include a GUI-style administrative interface. There are
user interfaces for adminBefore version 2.2, concurrency control was implemented third-party projects that offer
on a per-mongod basis. With version 2.2, concurrency
control was implemented at the database level.[23] Since
version 3.0,[24] pluggable storage engines were introduced, and each storage engine may implement concur- Licensing and support
rency control differently.[25] With MongoDB 3.0 concurfree under the GNU Affero
rency control is implemented at the collection level for MongoDB is available for
The language drivers are
the MMAPv1 storage engine,
and at the document
In addition, Mon[27]
level with the WiredTiger storage engine.
With vergoDB
sions prior to 3.0, one approach to increase concurrency
is to use sharding.[28] In some situations, reads and writes
will yield their locks. If MongoDB predicts a page is unlikely to be in memory, operations will yield their lock 9.2.5 Performance
while the pages load. The use of lock yielding expanded
United Software Associates published a benchmark usgreatly in 2.2.[29]
ing Yahoo's Cloud Serving Benchmark as a basis of all
Another criticism is related to the limitations of Monthe tests. MongoDB provides greater performance than
goDB when used on 32-bit systems. In some cases, this
Couchbase Server or Cassandra in all the tests they ran,
was due to inherent memory limitations.
in some cases by as much as 25x.[39]
recommends 64-bit systems and that users provide sufficient RAM for their working set. Compose, a provider of Another benchmark for top NoSQL databases utilizing
managed MongoDB infrastructure, recommends a scal- Amazon's Elastic Compute Cloud that was done by End
Point arrived at opposite results, placing MongoDB last
ing checklist for large systems.[32]
among the tested databases.[40]
Additionally, MongoDB does not support collation-based
sorting and is limited to byte-wise comparison via
memcmp,[33] which will not provide correct ordering for
many non-English languages[34] when used with a Uni- 9.2.6 Production deployments
code encoding.
Large-scale deployments of MongoDB are tracked by
MongoDB Inc. Notable users of MongoDB include:
Language support
MongoDB has official drivers for a variety of popular programming languages and development
environments.[35] There are also a large number of
unofficial or community-supported drivers for other
programming languages and frameworks.[36]
• Adobe: Adobe Experience Manager is intended to
accelerate development of digital experiences that
increase customer loyalty, engagement and demand.
Adobe uses MongoDB to store petabytes of data in
the large-scale content repositories underpinning the
Experience Manager.[41]
• Amadeus IT Group uses MongoDB for its back-end
• The Compact Muon Solenoid at CERN uses MongoDB as the primary back-end for the Data Aggregation System for the Large Hadron Collider.[43]
• HyperDex, a NoSQL database providing the MongoDB API with stronger consistency guarantees
• Craigslist: With 80 million classified ads posted 9.2.8 References
every month, Craigslist needs to archive billions
of records in multiple formats, and must be able [1] “Release Notes for MongoDB 3.0”. MongoDB.
to query and report on these archives at runtime.
Craigslist migrated from MySQL to MongoDB to [2] “10gen embraces what it created, becomes MongoDB
Inc.”. Gigaom. Retrieved 27 August 2013.
support its active archive, with continuous availability mandated for regulatory compliance across 700 [3] “Popularity ranking of database management systems”.
sites in 70 different countries.[44] Solid IT. Retrieved 2015-07-04.
• eBay uses MongoDB in the search suggestion and
the internal Cloud Manager State Hub.[45]
[4] “Release Notes for MongoDB 1.2.x”. Retrieved 2015-1129.
• FIFA (video game series): EA Sports FIFA is the
world’s best-selling sports video game franchise. To
serve millions of players, EA’s Spearhead development studio selected MongoDB[46] to store user data
and game state. Auto-sharding makes it simple to
scale MongoDB across EA’s 250+ servers with no
limits to growth as EA FIFA wins more fans.
[5] “Release Notes for MongoDB 1.4”. Retrieved 2015-1129.
• Foursquare deploys MongoDB on Amazon AWS to
store venues and user check-ins into venues.[47]
[8] “Release Notes for MongoDB 2.0”. Retrieved 2015-1129.
• LinkedIn uses MongoDB as its backend DB.[48]
[9] “Release Notes for MongoDB 2.2”. Retrieved 2015-1129.
• McAfee: MongoDB powers McAfee Global Threat
Intelligence (GTI), a cloud-based intelligence service that correlates data from millions of sensors around the globe. Billions of documents are
stored and analyzed in MongoDB to deliver realtime threat intelligence to other McAfee end-client
[6] “Release Notes for MongoDB 1.6”. Retrieved 2015-1129.
[7] “Release Notes for MongoDB 1.8”. Retrieved 2015-1129.
[10] “Release Notes for MongoDB 2.4”. Retrieved 2015-1129.
[11] “Release Notes for MongoDB 2.6”. Retrieved 2015-1129.
[12] “Release Notes for MongoDB 3.0”. Retrieved 2015-1129.
• MetLife uses MongoDB for “The Wall”, a customer
service application providing a “360-degree view” of [13] “Development Release Notes for 3.2 Release Candidate”.
Retrieved 2015-11-29.
MetLife customers.[50]
• SAP uses MongoDB in the SAP PaaS.[51]
[14] MongoDB. “MongoDB Developer Manual”. MongoDB.
• Shutterfly uses MongoDB for its photo platform. As
of 2013, the photo platform stores 18 billion photos
uploaded by Shutterfly’s 7 million users.[52]
[15] Data Modeling for MongoDB
• Tuenti uses MongoDB as its backend DB.[53]
[17] MongoDB. “Introduction to Sharding”. MongoDB.
• Yandex: The largest search engine in Russia uses
MongoDB to manage all user and metadata for its
file sharing service. MongoDB has scaled[54] to support tens of billions of objects and TBs of data,
growing at 10 million new file uploads per day.
[18] MongoDB. “GridFS article on MongoDB Developer’s
Manual”. MongoDB.
[16] MongoDB. “Introduction to Replication”. MongoDB.
[19] “NGINX plugin for MongoDB source code”. GitHub.
[20] “lighttpd plugin for MongoDB source code”. Bitbucket.
[21] Malick Md. “MongoDB overview”. Expertstown.
See also
• Server-side scripting
• MEAN, a solutions stack using MongoDB as the
[22] Kyle Kingsbury (2015-04-20). “Call me maybe: MongoDB stale reads”. Retrieved 2015-07-04.
[23] “MongoDB Jira Ticket 4328”.
[24] Eliot Horowitz (2015-01-22). “Renaming Our Upcoming
Release to MongoDB 3.0”. MongoDB. Retrieved 201502-23.
[25] “MongoDB 2.8 release”. MongoDB.
[26] MongoDB. “MMAPv1 Concurrency Improvement”.
[27] MongoDB. “WiredTiger Concurrency and Compression”.
[51] Richard Hirsch (30 September 2011). “The Quest to Understand the Use of MongoDB in the SAP PaaS”.
[52] Guy Harrison (28 January 2011). “Real World NoSQL:
MongoDB at Shutterfly”. Gigaom.
[53] “We host the MongoDB user group meetup at our office”.
[28] MongoDB. “FAQ Concurrency - How Does Sharding Affect Concurrency”. MongoDB.
[54] “Yandex: MongoDB”. Yandex.
[29] MongoDB. “FAQ Concurrency - Do Operations Ever
Yield the Lock”. MongoDB.
9.2.9 Bibliography
[30] MongoDB (8 July 2009). “32-bit Limitations”. MongoDB.
[31] David Mytton (25 September 2012). “Does Everybody
Hate MongoDB”. Server Density.
or empty |title= (help)
[33] “memcmp”. 31 May 2013. Retrieved
26 April 2014.
[34] “MongoDB Jira ticket 1920”.
[35] MongoDB. “MongoDB Drivers and Client Libraries”.
MongoDB. Retrieved 2013-07-08.
[36] MongoDB. “Community Supported Drivers”. MongoDB.
Retrieved 2014-07-09.
[37] MongoDB. “Admin UIs”. Retrieved 15 September 2015.
[38] MongoDB. “The AGPL”.
Database Blog. MongoDB.
• Hoberman, Steve (June 1, 2014), Data Modeling for
MongoDB (1st ed.), Technics Publications, p. 226,
ISBN 978-1-935504-70-2
• Banker, Kyle (March 28, 2011), MongoDB in Action
(1st ed.), Manning, p. 375, ISBN 978-1-93518287-0
• Chodorow, Kristina; Dirolf, Michael (September
23, 2010), MongoDB: The Definitive Guide (1st ed.),
O'Reilly Media, p. 216, ISBN 978-1-4493-8156-1
• Pirtle, Mitch (March 3, 2011), MongoDB for Web
Development (1st ed.), Addison-Wesley Professional, p. 360, ISBN 978-0-321-70533-4
• Hawkins, Tim; Plugge, Eelco; Membrey, Peter
(September 26, 2010), The Definitive Guide to MongoDB: The NoSQL Database for Cloud and Desktop
Computing (1st ed.), Apress, p. 350, ISBN 978-14302-3051-9
The MongoDB NoSQL
[39] United Software Associates. “High Performance Benchmarking: MongoDB and NoSQL Systems” (PDF).
[40] End Point (13 April 2015). “Benchmarking Top NoSQL
Databases; Apache Cassandra, Couchbase, HBase, and
MongoDB” (PDF).
[41] MongoDB. “Adobe Experience Manager”. MongoDB.
[42] “Presentation by Amadeus 11/2014”. MongoDB.
[43] “Holy Large Hadron Collider, Batman!". MongoDB.
[44] MongoDB. “Craigslist”. MongoDB.
[45] “MongoDB at eBay”. Slideshare.
[46] “MongoDB based FIFA Online”. MongoDB.
[47] “Experiences Deploying MongoDB on AWS”.
9.2.10 External links
• Official website
9.3 PostgreSQL
PostgreSQL, often simply Postgres, is an objectrelational database management system (ORDBMS) with
an emphasis on extensibility and standards-compliance.
As a database server, its primary function is to store data
securely, supporting best practices, and to allow for retrieval at the request of other software applications. It can
handle workloads ranging from small single-machine applications to large Internet-facing applications with many
concurrent users.
PostgreSQL implements the majority of the SQL:2011
standard,[9][10] is ACID-compliant and transactional (including most DDL statements) avoiding locking issues
[48] “Presentation by LinkedIn”. MongoDB.
using multiversion concurrency control (MVCC), pro[49] MongoDB. “McAfee is Improving Global Cybersecurity vides immunity to dirty reads and full serializability; hanwith MongoDB”. MongoDB.
dles complex SQL queries using many indexing meth[50] Doug Henschen (13 May 2013). “Metlife uses nosql ods that are not available in other databases; has upfor customer service”. Information Week. Retrieved 8 dateable views and materialized views, triggers, foreign
November 2014.
keys; supports functions and stored procedures, and other
expandability,[11] and has a large number of extensions
written by third parties. In addition to the possibility
of working with the major proprietary and open source
databases, PostgreSQL supports migration from them, by
its extensive standard SQL support and available migration tools. Proprietary extensions in databases such as
Oracle can be emulated by built-in and third-party open
source compatibility extensions. Recent versions also
provide replication of the database itself for availability
and scalability.
PostgreSQL is cross-platform and runs on many
operating systems including Linux, FreeBSD, OS X,
Solaris, and Microsoft Windows. On OS X, PostgreSQL
has been the default database starting with Mac OS
X 10.7 Lion Server, [12][13][14] and PostgreSQL client
tools are bundled with in the desktop edition. The
vast majority of Linux distributions have it available in
supplied packages.
PostgreSQL is developed by the PostgreSQL Global Development Group, a diverse group of many companies
and individual contributors.[15] It is free and open-source
software, released under the terms of the PostgreSQL License, a permissive free-software license.
PostgreSQL’s developers pronounce it /ˈpoʊstɡrɛs ˌkjuː
ˈɛl/.[16] It is abbreviated as Postgres, its original name.
Because of ubiquitous support for the SQL Standard
among most relational databases, the community considered changing the name back to Postgres. However,
the PostgreSQL Core Team announced in 2007 that the
product would continue to use the name PostgreSQL.[17]
The name refers to the project’s origins as a “post-Ingres"
database, being a development from University Ingres
DBMS (Ingres being an acronym for INteractive Graphics
Retrieval System).[18][19]
PostgreSQL evolved from the Ingres project at the
University of California, Berkeley. In 1982 the leader
of the Ingres team, Michael Stonebraker, left Berkeley
to make a proprietary version of Ingres.[18] He returned
to Berkeley in 1985, and started a post-Ingres project to
address the problems with contemporary database systems that had become increasingly clear during the early
1980s. The new project, POSTGRES, aimed to add the
fewest features needed to completely support types.[20]
These features included the ability to define types and to
fully describe relationships – something used widely before but maintained entirely by the user. In POSTGRES,
the database “understood” relationships, and could retrieve information in related tables in a natural way using
rules. POSTGRES used many of the ideas of Ingres, but
not its code.[21]
Starting in 1986, the POSTGRES team published a number of papers describing the basis of the system, and by
1987 had a prototype version shown at the 1988 ACM
SIGMOD Conference. The team released version 1 to
a small number of users in June 1989, then version 2
with a re-written rules system in June 1990. Version 3,
released in 1991, again re-wrote the rules system, and
added support for multiple storage managers and an improved query engine. By 1993, the great number of users
began to overwhelm the project with requests for support
and features. After releasing version 4.2[22] on June 30,
1994—primarily a cleanup—the project ended. Berkeley had released POSTGRES under an MIT-style license,
which enabled other developers to use the code for any
use. At the time, POSTGRES used an Ingres-influenced
POSTQUEL query language interpreter, which could be
interactively used with a console application named monitor.
In 1994, Berkeley graduate students Andrew Yu and Jolly
Chen replaced the POSTQUEL query language interpreter with one for the SQL query language, creating
Postgres95. The front-end program monitor was also replaced by psql. Yu and Chen released the code on the
On July 8, 1996, Marc Fournier at Networking
Services provided the first non-university development
server for the open-source development effort.[1] With the
participation of Bruce Momjian and Vadim B. Mikheev,
work began to stabilize the code inherited from Berkeley.
The first open-source version was released on August 1,
In 1996, the project was renamed to PostgreSQL to reflect
its support for SQL. The online presence at the website began on October 22, 1996.[23] The first
PostgreSQL release formed version 6.0 on January 29,
1997. Since then a group of developers and volunteers
around the world have maintained the software as The
PostgreSQL Global Development Group.
The PostgreSQL project continues to make major releases (approximately annually) and minor “bugfix” releases, all available under its free and open-source software PostgreSQL License. Code comes from contributions from proprietary vendors, support companies, and
open-source programmers at large. See also Release history below.
9.3.3 Multiversion concurrency control
PostgreSQL manages concurrency through a system
known as multiversion concurrency control (MVCC),
which gives each transaction a “snapshot” of the database,
allowing changes to be made without being visible to
other transactions until the changes are committed. This
largely eliminates the need for read locks, and ensures
the database maintains the ACID (atomicity, consistency,
isolation, durability) principles in an efficient manner.
PostgreSQL offers three levels of transaction isolation:
Read Committed, Repeatable Read and Serializable. Because PostgreSQL is immune to dirty reads, requesting
a Read Uncommitted transaction isolation level provides
read committed instead. Prior to PostgreSQL 9.1, requesting Serializable provided the same isolation level as
Repeatable Read. PostgreSQL 9.1 and later support full
serializability via the serializable snapshot isolation (SSI)
Storage and replication
There are also several asynchronous trigger-based replication packages for PostgreSQL. These remain useful
even after introduction of the expanded core capabilities, for situations where binary replication of an entire
database cluster is not the appropriate approach:
• Slony-I
• Londiste, part of SkyTools (developed by Skype)
• Bucardo multi-master replication (developed by[28]
• SymmetricDS multi-master, multi-tier replication
PostgreSQL, beginning with version 9.0, includes built-in
binary replication, based on shipping the changes (write- PostgreSQL includes built-in support for regular B-tree
and hash indexes, and two types of inverted indexes: genahead logs) to replica nodes asynchronously.
eralized search trees (GiST) and generalized inverted inVersion 9.0 also introduced the ability to run read-only dexes (GIN). Hash indexes are implemented, but discourqueries against these replicated nodes, where earlier ver- aged because they cannot be recovered after a crash or
sions would only allow that after promoting them to be a power loss. In addition, user-defined index methods can
new master. This allows splitting read traffic among mul- be created, although this is quite an involved process. Intiple nodes efficiently. Earlier replication software that dexes in PostgreSQL also support the following features:
allowed similar read scaling normally relied on adding
replication triggers to the master, introducing additional
• Expression indexes can be created with an index of
load onto it.
the result of an expression or function, instead of
Beginning from version 9.1, PostgreSQL also includes
simply the value of a column.
built-in synchronous replication[25] that ensures that, for
• Partial indexes, which only index part of a table, can
each write transaction, the master waits until at least one
be created by adding a WHERE clause to the end
replica node has written the data to its transaction log.
of the CREATE INDEX statement. This allows a
Unlike other database systems, the durability of a transsmaller index to be created.
action (whether it is asynchronous or synchronous) can
be specified per-database, per-user, per-session or even
• The planner is capable of using multiple indexes toper-transaction. This can be useful for work loads that
gether to satisfy complex queries, using temporary
do not require such guarantees, and may not be wanted
in-memory bitmap index operations.
for all data as it will have some negative effect on performance due to the requirement of the confirmation of the
• As of PostgreSQL 9.1, k-nearest neighbors (k-NN)
transaction reaching the synchronous standby.
indexing (also referred to KNN-GiST) provides effiThere can be a mixture of synchronous and asynchronous
standby servers. A list of synchronous standby servers
can be specified in the configuration which determines
which servers are candidates for synchronous replication.
The first in the list which is currently connected and actively streaming is the one that will be used as the current
synchronous server. When this fails, it falls to the next in
Synchronous multi-master replication is currently not included in the PostgreSQL core. Postgres-XC which
is based on PostgreSQL provides scalable synchronous
multi-master replication,[26] available in version 1.1 is licensed under the same license as PostgreSQL. A similar
project is called Postgres-XL and is available under the
Mozilla Public License.[27]
cient searching of “closest values” to that specified,
useful to finding similar words, or close objects or
locations with geospatial data. This is achieved without exhaustive matching of values.
• In PostgreSQL 9.2 and above, index-only scans often allow the system to fetch data from indexes without ever having to access the main table.
In PostgreSQL, all objects (with the exception of roles
and tablespaces) are held within a schema. Schemas
effectively act like namespaces, allowing objects of the
same name to co-exist in the same database.
The community has also written some tools to make man- By default, databases are created with the “public”
aging replication clusters easier, such as repmgr.
schema, but any additional schemas can be added, and the
public schema isn't mandatory. A “search_path” determines the order in which schemas are checked on unqualified objects (those without a prefixed schema), which can
be configured on a database or role level. The search path,
by default, contains the special schema name of "$user”,
which first looks for a schema named after the connected
database user (e.g. if the user “dave” were connected, it
would first look for a schema also named “dave” when referring to any objects). If such a schema is not found, it
then proceeds to the next schema. New objects are created in whichever valid schema (one that presently exists)
is listed first in the search path.
In addition, users can create their own data types which
can usually be made fully indexable via PostgreSQL’s
GiST infrastructure. Examples of these include the
geographic information system (GIS) data types from the
PostGIS project for PostgreSQL.
There is also a data type called a “domain”, which is the
same as any other data type but with optional constraints
defined by the creator of that domain. This means any
data entered into a column using the domain will have to
conform to whichever constraints were defined as part of
the domain.
Starting with PostgreSQL 9.2, a data type that represents
a range of data can be used which are called range types.
These can be discrete ranges (e.g. all integer values 1
Data types
to 10) or continuous ranges (e.g. any point in time beA wide variety of native data types are supported, includ- tween 10:00 am and 11:00 am). The built-in range types
available include ranges of integers, big integers, decimal
numbers, time stamps (with and without time zone) and
• Boolean
Custom range types can be created to make new types of
• Arbitrary precision numerics
ranges available, such as IP address ranges using the inet
type as a base, or float ranges using the float data type
• Character (text, varchar, char)
as a base. Range types support inclusive and exclusive
range boundaries using the [] and () characters respec• Binary
tively. (e.g. '[4,9)' represents all integers starting from
and including 4 up to but not including 9.) Range types
• Date/time (timestamp/time with/without timezone, are also compatible with existing operators used to check
date, interval)
for overlap, containment, right of etc.
• Money
• Enum
• Bit strings
• Text search type
• Composite
• HStore (an extension enabled key-value store within
• Arrays (variable length and can be of any data type,
including text and composite types) up to 1 GB in
total storage size
User-defined objects
New types of almost all objects inside the database can
be created, including:
• Casts
• Conversions
• Data types
• Domains
• Functions, including aggregate functions and window functions
• Geometric primitives
• Indexes including custom indexes for custom types
• IPv4 and IPv6 addresses
• Operators (existing ones can be overloaded)
• CIDR blocks and MAC addresses
• Procedural languages
• XML supporting XPath queries
Tables can be set to inherit their characteristics from a
• JSON (since version 9.2), and a faster binary “parent” table. Data in child tables will appear to exist in
JSONB (since version 9.4; not the same as the parent tables, unless data is selected from the parent
BSON[29] )
table using the ONLY keyword, i.e. SELECT * FROM
ONLY parent_table;. Adding a column in the parent table
will cause that column to appear in the child table.
Inheritance can be used to implement table partitioning,
using either triggers or rules to direct inserts to the parent
table into the proper child tables.
As of 2010, this feature is not fully supported yet—in
particular, table constraints are not currently inheritable.
All check constraints and not-null constraints on a parent
table are automatically inherited by its children. Other
types of constraints (unique, primary key, and foreign key
constraints) are not inherited.
Inheritance provides a way to map the features of generalization hierarchies depicted in Entity Relationship Diagrams (ERD) directly into the PostgreSQL database.
Other storage features
• DBD::Pg: Perl DBI driver
• JDBC: JDBC interface
• Lua: Lua interface
• Npgsql: .NET data provider
• ST-Links SpatialKit: Link Tool to ArcGIS
• PostgreSQL.jl: Julia interface
• node-postgres: Node.js interface
• pgoledb: OLEDB interface
• psqlODBC: ODBC interface
• psycopg2:[30] Python interface (also used by
• Referential integrity constraints including foreign
key constraints, column constraints, and row checks
• pgtclng: Tcl interface
• Binary and textual large-object storage
• pyODBC: Python library
• Tablespaces
• php5-pgsql: PHP driver based on libpq
• Per-column collation (from 9.1)
• postmodern: A Common Lisp interface
• Online backup
• pq: A pure Go PostgreSQL driver for the Go
database/sql package. The driver passes the compatibility test suite.[31]
• Point-in-time recovery, implemented using writeahead logging
• In-place upgrades with pg_upgrade for less down- Procedural languages
time (supports upgrades from 8.3.x and later)
Procedural languages allow developers to extend the
database with custom subroutines (functions), often
9.3.5 Control and connectivity
called stored procedures. These functions can be used
to build triggers (functions invoked upon modification of
Foreign data wrappers
certain data) and custom aggregate functions. Procedural
As of version 9.1, PostgreSQL can link to other systems languages can also be invoked without defining a functo retrieve data via foreign data wrappers (FDWs). These tion, using the “DO” command at SQL level.
can take the form of any data source, such as a file system, Languages are divided into two groups: “Safe” languages
another RDBMS, or a web service. This means regular are sandboxed and can be safely used by any user. Procedatabase queries can use these data sources like regular dures written in “unsafe” languages can only be created by
tables, and even join multiple data sources together.
superusers, because they allow bypassing the database’s
security restrictions, but can also access sources external
to the database. Some languages like Perl provide both
safe and unsafe versions.
PostgreSQL has several interfaces available and is also PostgreSQL has built-in support for three procedural lanwidely supported among programming language li- guages:
braries. Built-in interfaces include libpq (PostgreSQL’s
official C application interface) and ECPG (an embed• Plain SQL (safe). Simpler SQL functions can get
ded C system). External interfaces include:
expanded inline into the calling (SQL) query, which
saves function call overhead and allows the query optimizer to “see inside” the function.
• libpqxx: C++ interface
• PostgresDAC: PostgresDAC (for Embarcadero
RadStudio/Delphi/CBuilder XE-XE3)
• PL/pgSQL (safe), which resembles Oracle’s
PL/SQL procedural language and SQL/PSM.
• C (unsafe), which allows loading custom shared li- Npgsql, psycopg and node.js) so it can be used by external
braries into the database. Functions written in C of- applications.
fer the best performance, but bugs in code can crash
and potentially corrupt the database. Most built-in
functions are written in C.
In addition, PostgreSQL allows procedural languages to
be loaded into the database through extensions. Three
language extensions are included with PostgreSQL to
support Perl, Python and Tcl. There are external projects
to add support for many other languages, including Java,
JavaScript (PL/V8), R.
Rules allow the “query tree” of an incoming query to
be rewritten. Rules, or more properly, “Query Re-Write
Rules”, are attached to a table/class and “Re-Write” the
incoming DML (select, insert, update, and/or delete) into
one or more queries that either replace the original DML
statement or execute in addition to it. Query Re-Write
occurs after DML statement parsing, but before query
Other querying features
Triggers are events triggered by the action of SQL DML
statements. For example, an INSERT statement might
activate a trigger that checks if the values of the statement are valid. Most triggers are only activated by either
INSERT or UPDATE statements.
Triggers are fully supported and can be attached to tables.
In PostgreSQL 9.0 and above, triggers can be per-column
and conditional, in that UPDATE triggers can target specific columns of a table, and triggers can be told to execute under a set of conditions as specified in the trigger’s
WHERE clause. As of PostgreSQL 9.1, triggers can be
attached to views by utilising the INSTEAD OF condition. Views in versions prior to 9.1 can have rules, though.
Multiple triggers are fired in alphabetical order. In addition to calling functions written in the native PL/pgSQL,
triggers can also invoke functions written in other languages like PL/Python or PL/Perl.
Asynchronous notifications
PostgreSQL provides an asynchronous messaging system
that is accessed through the NOTIFY, LISTEN and UNLISTEN commands. A session can issue a NOTIFY
command, along with the user-specified channel and an
optional payload, to mark a particular event occurring.
Other sessions are able to detect these events by issuing a
LISTEN command, which can listen to a particular channel. This functionality can be used for a wide variety of
purposes, such as letting other sessions know when a table
has updated or for separate applications to detect when
a particular action has been performed. Such a system
prevents the need for continuous polling by applications
to see if anything has yet changed, and reducing unnecessary overhead. Notifications are fully transactional, in
that messages are not sent until the transaction they were
sent from is committed. This eliminates the problem of
messages being sent for an action being performed which
is then rolled back.
Many of the connectors for PostgreSQL provide support for this notification system (including libpq, JDBC,
• Transactions
• Full text search
• Views
• Materialized views[32]
• Updateable views[33]
• Recursive views[34]
• Inner, outer (full, left and right), and cross joins
• Sub-selects
• Correlated sub-queries[35]
• Regular expressions[36]
• Common table expressions and writable common table expressions
• Encrypted connections via TLS (current versions do
not use vulnerable SSL, even with that configuration
• Domains
• Savepoints
• Two-phase commit
• TOAST (The Oversized-Attribute Storage Technique)
is used to transparently store large table attributes
(such as big MIME attachments or XML messages)
in a separate area, with automatic compression.
• Embedded SQL is implemented using preprocessor. SQL code is first written embedded into C
code. Then code is run through ECPG preprocessor,
which replaces SQL with calls to code library. Then
code can be compiled using a C compiler. Embedding works also with C++ but it does not recognize
all C++ constructs.
PostgreSQL manages its internal security on a per-role
basis. A role is generally regarded to be a user (a role
that can log in), or a group (a role of which other roles
are members). Permissions can be granted or revoked
on any object down to the column level, and can also allow/prevent the creation of new objects at the database,
schema or table levels.
The sepgsql extension (provided with PostgreSQL as of
version 9.1) provides an additional layer of security by integrating with SELinux. This utilises PostgreSQL’s SECURITY LABEL feature.
PostgreSQL natively supports a broad number of external
authentication mechanisms, including:
• password (either MD5 or plain-text)
• Kerberos
• ident (maps O/S user-name as provided by an ident
server to database user-name)
• peer (maps local user name to database user name)
• Active Directory
• certificate
The GSSAPI, SSPI, Kerberos, peer, ident and certificate
methods can also use a specified “map” file that lists which
users matched by that authentication system are allowed
to connect as a specific database user.
These methods are specified in the cluster’s host-based
authentication configuration file (pg_hba.conf), which
determines what connections are allowed. This allows
control over which user can connect to which database,
where they can connect from (IP address/IP address
range/domain socket), which authentication system will
be enforced, and whether the connection must use TLS.
Upcoming features
Upcoming features in 9.5, in order of commit, include:
• IMPORT FOREIGN SCHEMA to import foreign
tables from a foreign schema, meaning tables no
longer have to be manually configured[38]
for switching tables between logged and unlogged
• Row-Level Security Policies for controlling which
rows are visible or can be added to a table[40]
• SKIP LOCKED for row-level locks[41]
• BRIN (Block Range Indexes) to speed up queries on
very large tables[42]
• Parallel VACUUMing with vacuumdb tool[43]
• Foreign tables can inherit and be inherited from[44]
• pg_rewind tool to efficiently resynchronise failed
primary to new primary[45]
• Index-only scans on GiST indexes[46]
• CREATE TRANSFORM for mapping data type
structures to procedural language data types[47]
• INSERT ...
• JSONB-modifying operators and functions[49]
Upcoming features in 9.6, in order of commit, include:
• Parallel sequential scan[52]
9.3.8 Add-ons
• MADlib: an open source analytics library for
PostgreSQL providing mathematical, statistical and
machine-learning methods for structured and unstructured data
• MySQL migration wizard: included with EnterpriseDB’s PostgreSQL installer (source code also
• Performance Wizard: included with EnterpriseDB’s
PostgreSQL installer (source code also available)[53]
• pgRouting: extended PostGIS to provide geospatial
routing functionality[54] (GNU GPL)
• PostGIS: a popular add-on which provides support
for geographic objects (GNU GPL)
• Postgres Enterprise Manager: a non-free tool consisting of a service, multiple agents, and a GUI
which provides remote monitoring, management,
reporting, capacity planning and tuning[55]
• ST-Links SpatialKit: Extension for directly connecting to spatial databases[56]
Benchmarks and performance
Univel. Most other Unix-like systems should also work.
PostgreSQL works on any of the following instruction
set architectures: x86 and x86-64 on Windows and
other operating systems; these are supported on other
than Windows: IA-64 Itanium, PowerPC, PowerPC 64,
S/390, S/390x, SPARC, SPARC 64, Alpha, ARMv8A (64-bit)[67] and older ARM (32-bit, including older
such as ARMv6 in Raspberry Pi[68] ), MIPS, MIPSel,
M68k, and PA-RISC. It is also known to work on M32R,
The first industry-standard and peer-validated benchmark NS32k, and VAX. In addition to these, it is possible to
was completed in June 2007 using the Sun Java System build PostgreSQL for an unsupported CPU by disabling
Application Server (proprietary version of GlassFish) 9.0 spinlocks.[69]
Platform Edition, UltraSPARC T1-based Sun Fire server
and PostgreSQL 8.2.[59] This result of 778.14 SPECjAppServer2004 [email protected] compares favourably 9.3.11 Database administration
with the 874 [email protected] with Oracle 10 on an
Itanium-based HP-UX system.[57]
See also: Comparison of database tools
Many informal performance studies of PostgreSQL have
been done.[57] Performance improvements aimed at improving scalability started heavily with version 8.1. Simple benchmarks between version 8.0 and version 8.4
showed that the latter was more than 10 times faster on
read-only workloads and at least 7.5 times faster on both
read and write workloads.[58]
In August 2007, Sun submitted an improved benchmark
score of 813.73 SPECjAppServer2004 [email protected] Open source front-ends and tools for administering PostWith the system under test at a reduced price, the greSQL include:
price/performance improved from $US 84.98/JOPS to
$US 70.57/JOPS.[60]
psql The primary front-end for PostgreSQL is the psql
The default configuration of PostgreSQL uses only a
command-line program, which can be used to ensmall amount of dedicated memory for performanceter SQL queries directly, or execute them from a
critical purposes such as caching database blocks and
file. In addition, psql provides a number of metasorting. This limitation is primarily because older opercommands and various shell-like features to faciliating systems required kernel changes to allow allocating
tate writing scripts and automating a wide variety of
large blocks of shared memory.[61] protasks; for example tab completion of object names
vides advice on basic recommended performance pracand SQL syntax.
tice in a wiki.[62]
In April 2012, Robert Haas of EnterpriseDB demon- pgAdmin The pgAdmin package is a free and open
source graphical user interface administration tool
strated PostgreSQL 9.2’s linear CPU scalability using a
for PostgreSQL, which is supported on many comserver with 64 cores.
puter platforms.[70] The program is available in more
Matloob Khushi performed benchmarking between Postthan a dozen languages. The first prototype, named
gresql 9.0 and MySQL 5.6.15 for their ability to propgManager, was written for PostgreSQL 6.3.2 from
cess genomic data. In his performance analysis he found
1998, and rewritten and released as pgAdmin unthat PostgreSQL extracts overlapping genomic regions 8x
der the GNU General Public License (GPL) in later
times faster than MySQL using two datasets of 80,000
months. The second incarnation (named pgAdmin
each forming random human DNA regions. Insertion and
II) was a complete rewrite, first released on January
data uploads in PostgreSQL were also better, although
16, 2002. The third version, pgAdmin III, was origgeneral searching capability of both databases was almost
inally released under the Artistic License and then
equivalent . [64]
released under the same license as PostgreSQL. Unlike prior versions that were written in Visual Basic,
pgAdmin III is written in C++, using the wxWidgets
9.3.10 Platforms
framework allowing it to run on most common operating systems. The query tool includes a scriptPostgreSQL is available for the following operating sysing language called pgScript for supporting admin
tems: Linux (all recent distributions), Windows (Winand development tasks. In December 2014, Dave
dows 2000 SP4 and later) (compilable by e.g. Visual
Page, the pgAdmin project founder and primary
Studio, now with up to most recent 2015 version),
developer,[71] announced that with the shift towards
DragonFly BSD, FreeBSD, OpenBSD, NetBSD, Mac OS
web-based models work has started on pgAdmin 4
X,[14] AIX, BSD/OS, HP-UX, IRIX, OpenIndiana,[65]
with the aim of facilitating Cloud deployments.[72]
OpenSolaris, SCO OpenServer, SCO UnixWare, Solaris
Although still at the concept stage,[73] the plan is to
and Tru64 Unix. In 2012, support for the following obbuild a single Python-based pgAdmin that users can
solete systems was removed (still supported in 9.1):[66]
either deploy on a web server or run from their deskDG/UX, NeXTSTEP, SunOS 4, SVR4, Ultrix 4, and
phpPgAdmin phpPgAdmin is a web-based administration tool for PostgreSQL written in PHP and based
on the popular phpMyAdmin interface originally
written for MySQL administration.[74]
PostgreSQL Studio PostgreSQL Studio allows users to
perform essential PostgreSQL database development tasks from a web-based console. PostgreSQL
Studio allows users to work with cloud databases
without the need to open firewalls.[75]
TeamPostgreSQL AJAX/JavaScript-driven web interface for PostgreSQL. Allows browsing, maintaining
and creating data and database objects via a web
browser. The interface offers tabbed SQL editor
with auto-completion, row-editing widgets, clickthrough foreign key navigation between rows and
tables, 'favorites’ management for commonly used
scripts, among other features. Supports SSH for
both the web interface and the database connections. Installers are available for Windows, Mac and
Linux, as well as a simple cross-platform archive
that runs from a script.[76]
LibreOffice/ Base
LibreOffice/ Base can be used
as a front-end for PostgreSQL.[77][78]
pgFouine The pgFouine PostgreSQL log analyzer generates detailed reports from a PostgreSQL log file
and provides VACUUM analysis.[79]
A number of companies offer proprietary tools for PostgreSQL. They often consist of a universal core that is
adapted for various specific database products. These
tools mostly share the administration features with the
open source tools but offer improvements in data modeling, importing, exporting or reporting.
Prominent users
Prominent organizations and products that use PostgreSQL as the primary database include:
• Yahoo! for web user behavioral analysis, storing
two petabytes and claimed to be the largest data
warehouse using a heavily modified version of PostgreSQL with an entirely different column-based
storage engine and different query processing layer.
While for performance, storage, and query purposes the database bears little resemblance to PostgreSQL, the front-end maintains compatibility so
that Yahoo can use many off-the-shelf tools already
written to interact with PostgreSQL.[80][81]
• In 2009, social networking website MySpace
used Aster Data Systems's nCluster database for
data warehousing, which was built on unmodified
• uses PostgreSQL for their main genealogy
• OpenStreetMap, a collaborative project to create a
free editable map of the world.[85]
• Afilias, domain registries for .org, .info and
• Sony Online multiplayer online games.[87]
• BASF, shopping platform for their agribusiness
• Reddit social news website.[89]
• Skype VoIP
• Sun xVM, Sun’s virtualization and datacenter automation suite.[91]
• MusicBrainz, open online music encyclopedia.[92]
• The International Space Station for collecting
telemetry data in orbit and replicating it to the
• MyYearbook social networking site.[94]
• Instagram,
a popular mobile photo sharing
• Disqus, an online discussion and commenting
• TripAdvisor, travel information website of mostly
user-generated content[97]
PostgreSQL is offered by some major vendors as software
as a service:
• Heroku, a platform as a service provider, has supported PostgreSQL since the start in 2007.[98] They
offer value-add features like full database “rollback” (ability to restore a database from any point
in time),[99] which is based on WAL-E, open source
software developed by Heroku.[100]
• In January 2012, EnterpriseDB released a cloud version of both PostgreSQL and their own proprietary
Postgres Plus Advanced Server with automated provisioning for failover, replication, load-balancing,
and scaling. It runs on Amazon Web Services.[101]
• VMware offers vFabric Postgres for private clouds
on vSphere since May 2012.[102]
• In November 2013, announced that
they are adding PostgreSQL to their Relational
Database Service offering.[103][104]
Proprietary derivatives and support and provide professional support for open source PostgreSQL.
Although the license allows proprietary products based
on Postgres, the code did not develop in the proprietary
space at first. The first main offshoot originated when
Paula Hawthorn (an original Ingres team member who
moved from Ingres) and Michael Stonebraker formed
Illustra Information Technologies to make a proprietary
product based on POSTGRES.
Many other companies have used PostgreSQL as the
base for their proprietary database projects, e.g. Truviso, Netezza, ParAccel (used in Amazon Redshift[112] ).
In many cases the products have been enhanced so much
that the software has been forked, though with some features cherry-picked from later releases.
In 2000, former Red Hat investors created the company 9.3.14 Release history
Great Bridge to make a proprietary product based on
PostgreSQL and compete against proprietary database 9.3.15 See also
vendors. Great Bridge sponsored several PostgreSQL
developers and donated many resources back to the
• Comparison of relational database management syscommunity,[105] but by late 2001 closed due to tough
competition from companies like Red Hat and to poor
market conditions.[106][107]
In 2001, Command Prompt, Inc. released Mammoth 9.3.16 References
PostgreSQL, a proprietary product based on PostgreSQL.
[1] “Happy Birthday, PostgreSQL!". PostgreSQL Global DeIn 2008, Command Prompt, Inc. released the source unvelopment Group. July 8, 2008.
der the original license. Command Prompt, Inc. continues to support the PostgreSQL community actively [2] “2015-10-08 Security Update Release”. PostgreSQL. The
PostgreSQL Global Development Group. 2015-10-08.
through developer sponsorships and projects including
Retrieved 2015-10-08.
PL/Perl, PL/php, and hosting of community projects such
as the PostgreSQL build farm.
[3] “PostgreSQL licence approved by OSI”. Crynwr. 2010In January 2005, PostgreSQL received backing by
database vendor Pervasive Software, known for its
Btrieve product which was ubiquitous on the Novell NetWare platform. Pervasive announced commercial support and community participation and achieved some success. In July 2006, Pervasive left the PostgreSQL support
02-18. Retrieved 2010-02-18.
[4] “OSI PostgreSQL Licence”. Open Source Initiative.
2010-02-20. Retrieved 2010-02-20.
[5] “License”. PostgreSQL Global Development Group. Retrieved 2010-09-20.
[6] “Debian -- Details of package postgresql in sid”.
In mid-2005, two other companies announced plans to
make proprietary products based on PostgreSQL with focus on separate niche markets. EnterpriseDB added func- [7]
tionality to allow applications written to work with Oracle [8]
to be more readily run with PostgreSQL. Greenplum contributed enhancements directed at data warehouse and [9]
business intelligence applications, including the BizGres
In October 2005, John Loiacono, executive vice president
of software at Sun Microsystems, commented: “We're
not going to OEM Microsoft but we are looking at Post- [11]
greSQL right now,”[109] although no specifics were released at that time. By November 2005, Sun had announced support for PostgreSQL.[110] By June 2006, Sun
Solaris 10 (June 2006 release) shipped with PostgreSQL. [12]
"Licensing:Main". FedoraProject.
“SQL Conformance”. 2013-04-04. Retrieved 2013-08-28.
“Appendix D. SQL Conformance”. PostgreSQL 9 Documentation. PostgreSQL Global Development Group.
2009 [1996]. Retrieved 2013-04-01.
“What is PostgreSQL?". PostgreSQL 9.3.0 Documentation. PostgreSQL Global Development Group. Retrieved
“Lion Server: MySQL not included”. 2011-08-04. Retrieved 2011-11-12.
In August 2007, EnterpriseDB introduced Postgres
Plus (originally called EnterpriseDB Postgres), a pre- [13] “OS X Lion Server — Technical Specifications”. 201108-04. Retrieved 2011-11-12.
configured distribution of PostgreSQL including many
contrib modules and add-on components.
In 2011, 2ndQuadrant became a Platinum Sponsor of
[15] “Contributor Profiles”. PostgreSQL. Retrieved December
PostgreSQL, in recognition of their long-standing con17, 2011.
tributions and developer sponsorship. 2ndQuadrant employ one of the largest teams of PostgreSQL contributors [16] Audio sample, 5.6k MP3
[17] “Project name — statement from the core team”. 2007-11-16. Retrieved 2007-1116.
[18] Stonebraker, M; Rowe, LA (May 1986). The design of
POSTGRES (PDF). Proc. 1986 ACM SIGMOD Conference on Management of Data. Washington, DC. Retrieved 2011-12-17.
[20] Stonebraker, M; Rowe, LA. The POSTGRES data model
(PDF). Proceedings of the 13th International Conference
on Very Large Data Bases. Brighton, England: Morgan
Kaufmann Publishers. pp. 83–96. ISBN 0-934613-46X.
[38] “Implement IMPORT FOREIGN SCHEMA.”. 2014-0710. Retrieved 2014-09-11.
[39] “Implement ALTER TABLE .. SET LOGGED / UNLOGGED”. 2014-08-22. Retrieved 2014-08-27.
[40] “Row-Level Security Policies (RLS)". 2014-09-19. Retrieved 2014-09-19.
[41] “Implement SKIP LOCKED for row-level locks”. 201410-07. Retrieved 2014-10-07.
[42] “BRIN: Block Range Indexes”. 2014-11-07. Retrieved
[43] “vacuumdb: enable parallel mode”. 2015-01-23. Retrieved 2015-01-28.
[21] Pavel Stehule (9 June 2012). “Historie projektu PostgreSQL” (in Czech).
[44] “Allow foreign tables to participate in inheritance.”. 201503-22. Retrieved 2015-03-26.
[22] “University POSTGRES, Version 4.2”. 1999-07-26.
[45] “Add pg_rewind, for re-synchronizing a master server after failback.”. 2015-03-23. Retrieved 2015-03-26.
[23] Page, Dave (2015-04-07). “Re: 20th anniversary of PostgreSQL ?". pgsql-advocacy (Mailing list). Retrieved 9
April 2015.
[24] Dan R. K. Ports, Kevin Grittner (2012). “Serializable
Snapshot Isolation in PostgreSQL” (PDF). Proceedings of
the VLDB Endowment 5 (12): 1850–1861.
[25] PostgreSQL 9.1 with synchronous replication (news), H
[26] Postgres-XC project page (website), Postgres-XC
[46] “Add support for index-only scans in GiST.”. 2015-0326. Retrieved 2015-03-26.
[47] “Add transforms feature”. 2015-04-26. Retrieved 201504-27.
[48] “Add support for INSERT ... ON CONFLICT DO
NOTHING/UPDATE”. 2015-05-08. Retrieved 201505-11.
[49] “Additional functions and operators for jsonb”. 2015-0512. Retrieved 2015-05-16.
[27] Postgres-XL product page (website), TransLattice
[28] Marit Fischer (2007-11-10). “ finally
gives something back to the open source community”
(Press release).
[29] Geoghegan, Peter (March 23, 2014). “What I think of
[30] “PostgreSQL + Python — Psycopg”.
[31] “SQL database drivers”. Go wiki. Retrieved
22 June 2015.
[32] “Add a materialized view relations.”. 2013-03-04. Retrieved 2013-03-04.
[33] “Support automatically-updatable views.”. 2012-12-08.
Retrieved 2012-12-08.
[34] “Add CREATE RECURSIVE VIEW syntax”. 2013-0201. Retrieved 2013-02-28.
[35] Momjian, Bruce (2001). “Subqueries”. PostgreSQL: Introduction and Concepts. Addison-Wesley. ISBN 0-20170331-9. Retrieved 2010-09-25.
[36] Bernier, Robert (2 February 2006). “Using Regular Expressions in PostgreSQL”. O'Reilly Media. Retrieved
[37] “A few short notes about PostgreSQL and POODLE”.
[50] “TABLESAMPLE, SQL Standard and extensible”. 201505-15. Retrieved 2015-05-16.
[51] “Support GROUPING SETS, CUBE and ROLLUP.”.
2015-05-16. Retrieved 2015-05-16.
[52] “Generate parallel sequential scan plans in simple cases.”.
2015-11-11. Retrieved 2015-11-18.
[53] “Postgres Plus Downloads”. Company website. EnterpriseDB. Retrieved November 12, 2011.
[54] pgRouting, PostLBS
[55] “Postgres Enterprise Manager”. Company website. EnterpriseDB. Retrieved November 12, 2011.
[56] ST Links
[57] Josh Berkus (2007-07-06). “PostgreSQL publishes first
real benchmark”. Retrieved 2007-07-10.
[58] György Vilmos (2009-09-29).
Retrieved 2010-08-28.
“PostgreSQL history”.
[59] “SPECjAppServer2004 Result”. SPEC. 2007-07-06. Retrieved 2007-07-10.
[60] “SPECjAppServer2004 Result”. SPEC. 2007-07-04. Retrieved 2007-09-01.
[61] “Managing Kernel Resources”. PostgreSQL Manual. Retrieved November 12, 2011.
[62] Greg Smith, Robert Treat, and Christopher Browne.
“Tuning your PostgreSQL server”.
Wiki. Retrieved November 12, 2011.
[84] “Last Weekend’s Outage”. Blog. Geni. 2011-08-01.
[63] Robert Haas (2012-04-03). “Did I Say 32 Cores? How
about 64?". Retrieved 2012-04-08.
[86] PostgreSQL affiliates .ORG domain, AU: Computer World
[64] Matloob Khushi “Benchmarking database performance
for genomic data.”, J Cell Biochem., 2015 Jun;116(6):87783. doi: 10.1002/jcb.25049.
[65] “oi_151a Release Notes”. OpenIndiana. Retrieved 201204-07.
[66] “Git — postgresql.git/commitdiff".
Retrieved 2012-07-08.
[67] “AArch64 planning BoF at DebConf”.
[85] “Database”. Wiki. OpenStreetMap.
[87] Sony Online opts for open-source database over Oracle,
Computer World
[88] A Web Commerce Group Case Study on PostgreSQL (PDF)
(1.2 ed.), PostgreSQL
[89] “Architecture Overview”. Reddit software wiki. Reddit.
27 March 2014. Retrieved 2014-11-25.
[90] “PostgreSQL at Skype”. Skype Developer Zone. 2006.
Retrieved 2007-10-23.
[91] “How Much Are You Paying For Your Database?". Sun
step-5-update-installing-postgresql-on-my-raspberry-pi-1-and-2/Microsystems blog. 2007. Retrieved 2007-12-14.
[69] “Supported Platforms”. PostgreSQL Global Development
Group. Retrieved 2012-04-06.
[92] “Database — MusicBrainz”. MusicBrainz Wiki. Retrieved 5 February 2011.
[70] “pgAdmin: PostgreSQL administration and management
tools”. website. Retrieved November 12, 2011.
[93] Duncavage, Daniel P (2010-07-13).
Postgres-Nagios help”.
[71] “pgAdmin Development Team”. Retrieved
22 June 2015.
[94] Roy, Gavin M (2010).
“PostgreSQL at” (talk). USA East: PostgreSQL Conference.
[72] Dave, Page. “The story of pgAdmin”. Dave’s Postgres
Blog. Retrieved 7 December
[95] “Keeping Instagram up with over a million new users in
twelve hours”. 201105-17. Retrieved 2012-07-07.
[73] Dave, Page. “pgAdmin 4”. Git Repository - The next generation of pgAdmin.
[96] “Postgres at Disqus”. Retrieved May 24, 2013.
[74] phpPgAdmin Project (2008-04-25). “About phpPgAdmin”. Retrieved 2008-04-25.
[97] Matthew Kelly (27 March 2015). At The Heart Of
A Giant: Postgres At TripAdvisor. PGConf US 2015.
(Presentation video)
[75] PostgreSQL Studio (2013-10-09). “About PostgreSQL
Studio”. Retrieved 2013-10-09.
“NASA needs
[98] Alex Williams (1 April 2013). “Heroku Forces Customer
Upgrade To Fix Critical PostgreSQL Security Hole”.
[77] (2010-01-10). “Back Ends for OpenOffice”. Retrieved 2011-01-05.
[99] Barb Darrow (11 November 2013). “Heroku gussies up
Postgres with database roll-back and proactive alerts”. GigaOM.
[76] “TeamPostgreSQL website”.
[78] (2012-10-14). “Base features”. Retrieved [100] Craig Kerstiens (26 September 2013). “WAL-E and Con2012-10-14.
tinuous Protection with Heroku Postgres”. Heroku blog.
[79] Greg Smith (15 October 2010). PostgreSQL 9.0 High Per[101] “EnterpriseDB Offers Up Postgres Plus Cloud Database”.
formance. Packt Publishing. ISBN 978-1-84951-030-1. 2012-01-27. Retrieved 201207-07.
[80] Eric Lai (2008-05-22). “Size matters: Yahoo claims
2-petabyte database is world’s biggest,
[102] Al Sargent (15 May 2012). “Introducing VMware vFabric Suite 5.1: Automated Deployment, New Components,
and Open Source Support”. VMware blogs.
[81] Thomas Claburn (2008-05-21). “Yahoo Claims Record
With Petabyte Database”. InformationWeek.
[103] Jeff (14 November 2013). “Amazon RDS for PostgreSQL
[82] Emmanuel Cecchet (May 21, 2009). Building PetaByte
— Now Available”. Amazon Web Services Blog.
Warehouses with Unmodified PostgreSQL (PDF). PGCon
[104] Alex Williams (14 November 2013). “PostgreSQL Now
2009. Retrieved November 12, 2011.
Available On Amazon’s Relational Database Service”.
[83] “ scales analytics for all their friends”
(PDF). case study. Aster Data. June 15, 2010. Archived
(PDF) from the original on November 14, 2010. Re- [105] Maya Tamiya (2001-01-10).
“Interview: Bruce
trieved November 12, 2011.
Momjian”. Retrieved 2007-09-07.
[106] “Great Bridge ceases operations” (Press release). Great
Bridge. 2001-09-06. Retrieved 2007-09-07.
[107] Nikolai Bezroukov (1 July 2004). “The Sunset of Linux
Hype”. Portraits of Open Source Pioneers. NORFOLK,
Va., September 6, 2001 -- Great Bridge LLC, the company that pioneered commercial distribution and support
of the PostgreSQL open source database, announced today that it has ceased business operations
[108] John Farr (2006-07-25). “Open letter to the PostgreSQL
Community”. Pervasive Software. Archived from the
original on 2007-02-25. Retrieved 2007-02-13.
[109] Rodney Gedda (2005-10-05). “Sun’s software chief eyes
databases, groupware”. Computerworld. Retrieved 200702-13.
9.3.18 External links
• Official website
• PostgreSQL wiki
• – PostgreSQL’s
explain analyze made readable
• PostgreSQL at DMOZ
9.4 Apache Cassandra
[110] “Sun Announces Support for Postgres Database on Solaris
10” (Press release). Sun Microsystems. 2005-11-17. Retrieved 2007-02-13.
[111] “EnterpriseDB Announces First-Ever Professional-Grade
PostgreSQL Distribution for Linux” (Press release).
EnterpriseDB. 2007-08-07. Retrieved 2007-08-07.
[113] “Versioning policy”. PostgreSQL Global Development
Group. Retrieved 2012-01-30.
Helenos is a graphical user interface for Cassandra
Further reading
Apache Cassandra is an open source distributed
• Obe, Regina; Hsu, Leo (July 8, 2012). PostgreSQL: database management system designed to handle large
amounts of data across many commodity servers, providUp and Running. O'Reilly. ISBN 1-4493-2633-1.
ing high availability with no single point of failure. Cas• Krosing, Hannu; Roybal, Kirk (June 15, 2013). sandra offers robust support for clusters spanning multiPostgreSQL Server Programming. Packt Publishing. ple datacenters,[1] with asynchronous masterless replicaISBN 9781849516983.
tion allowing low latency operations for all clients.
• Riggs, Simon; Krosing, Hannu (October 27, 2010). Cassandra also places a high value on performance. In
PostgreSQL 9 Administration Cookbook. Packt Pub- 2012, University of Toronto researchers studying NoSQL
systems concluded that “In terms of scalability, there is
lishing. ISBN 1-84951-028-8.
a clear winner throughout our experiments. Cassandra
• Smith, Greg (October 15, 2010). PostgreSQL 9 High achieves the highest throughput for the maximum number
Performance. Packt Publishing. ISBN 1-84951- of nodes in all experiments” although “this comes at the
price of high write and read latencies.”[2]
• Gilmore, W. Jason; Treat, Robert (February 27,
2006). Beginning PHP and PostgreSQL 8: From 9.4.1 History
Novice to Professional. Apress. p. 896. ISBN 159059-547-5.
Apache Cassandra was initially developed at Facebook to
power their Inbox Search feature by Avinash Lakshman
• Douglas, Korry (August 5, 2005). PostgreSQL (Sec(one of the authors of Amazon’s Dynamo) and Prashant
ond ed.). Sams. p. 1032. ISBN 0-672-32756-2.
Malik. It was released as an open source project on
• Matthew, Neil; Stones, Richard (April 6, 2005). Google code in July 2008. In March 2009, it became
Beginning Databases with PostgreSQL (Second ed.). an Apache Incubator project. On February 17, 2010 it
graduated to a top-level project.[5]
Apress. p. 664. ISBN 1-59059-478-9.
• Worsley, John C; Drake, Joshua D (January 2002).
Practical PostgreSQL. O'Reilly Media. p. 636.
ISBN 1-56592-846-6.
It was named after the Greek mythological prophet
Releases after graduation include
• 0.6, released Apr 12 2010, added support for inte- Scalability Read and write throughput both increase lingrated caching, and Apache Hadoop MapReduce[7]
early as new machines are added, with no downtime
or interruption to applications.
• 0.7, released Jan 08 2011, added secondary indexes
and online schema changes[8]
Fault-tolerant Data is automatically replicated to mul• 0.8, released Jun 2 2011, added the Cassandra
tiple nodes for fault-tolerance. Replication across
Query Language (CQL), self-tuning memtables,
multiple data centers is supported. Failed nodes can
and support for zero-downtime upgrades[9]
be replaced with no downtime.
• 1.0, released Oct 17 2011, added integrated compression, leveled compaction, and improved read Tunable consistency Writes and reads offer a tunable
level of consistency, all the way from “writes never
fail” to “block for all replicas to be readable”, with
• 1.1, released Apr 23 2012, added self-tuning
the quorum level in the middle.[16]
caches, row-level isolation, and support for mixed
ssd/spinning disk deployments[11]
MapReduce support Cassandra has Hadoop integration, with MapReduce support. There is support
• 1.2, released Jan 2 2013, added clustering across
also for Apache Pig and Apache Hive.[17]
virtual nodes, inter-node communication, atomic
batches, and request tracing
• 2.0, released Sep 4 2013, added lightweight transac- Query language Cassandra
(Cassandra Query Language), a SQL-like altertions (based on the Paxos consensus protocol), trignative to the traditional RPC interface. Language
gers, improved compactions
drivers are available for Java (JDBC), Python
• 2.0.4, released Dec 30 2013, added allowing spec(DBAPI2), Node.JS (Helenus), Go (gocql) and
ifying datacenters to participate in a repair, client
encryption support to sstableloader, allow removing
snapshots of no-longer-existing CFs[13]
Below an example of keyspace creation, including a col[14]
umn family in CQL 3.0:[19]
• 2.1.0 released Sep 10 2014
CREATE KEYSPACE MyKeySpace WITH REPLICA• 2.1.6 released June 8, 2015
TION = { 'class’ : 'SimpleStrategy', 'replication_factor' :
• 2.1.7 released June 22, 2015
MyColumns (id text, Last text, First text, PRIMARY
• 2.2.0 released July 20, 2015
KEY(id)); INSERT INTO MyColumns (id, Last, First)
VALUES ('1', 'Doe', 'John'); SELECT * FROM My• 2.2.2 released October 5, 2015
Licensing and support
Which gives:
Apache Cassandra is an Apache Software Foundation id | first | last ----+-------+------ 1 | John | Doe (1 rows)
project, so it has an Apache License (version 2.0).
Main features
Decentralized Every node in the cluster has the same
role. There is no single point of failure. Data is
distributed across the cluster (so each node contains
different data), but there is no master as every node
can service any request.
9.4.4 Data model
Cassandra is essentially a hybrid between a key-value and
a column-oriented (or tabular) database. Its data model is
a partitioned row store with tunable consistency.[16] Rows
are organized into tables; the first component of a table’s
primary key is the partition key; within a partition, rows
are clustered by the remaining columns of the key.[20]
Supports replication and multi data center replication Other columns may be indexed separately from the priReplication strategies are configurable.[15] Cas- mary key.[21]
sandra is designed as a distributed system, for
at run-time
deployment of large numbers of nodes across Tables may be created, dropped, and altered
multiple data centers. Key features of Cassandra’s
distributed architecture are specifically tailored for Cassandra does not support joins or subqueries. Rather,
multiple-data center deployment, for redundancy, Cassandra emphasizes denormalization through features
like collections.[23]
for failover and disaster recovery.
A column family (called “table” since CQL 3) resembles a table in an RDBMS. Column families contain rows
and columns. Each row is uniquely identified by a row
key. Each row has multiple columns, each of which has
a name, value, and a timestamp. Unlike a table in an
RDBMS, different rows in the same column family do
not have to share the same set of columns, and a column
may be added to one or multiple rows at any time.[24]
Each key in Cassandra corresponds to a value which is an
object. Each key has values as columns, and columns are
grouped together into sets called column families. Thus,
each key identifies a row of a variable number of elements. These column families could be considered then
as tables. A table in Cassandra is a distributed multi dimensional map indexed by a key. Furthermore, applications can specify the sort order of columns within a Super
Column or Simple Column family.
When the cluster for Apache Cassandra is designed, an
important point is to select the right partitioner. Two partitioners exist:[25]
1. RandomPartitioner (RP): This partitioner randomly
distributes the key-value pairs over the network, resulting in a good load balancing. Compared to OPP,
more nodes have to be accessed to get a number of
2. OrderPreservingPartitioner (OPP): This partitioner
distributes the key-value pairs in a natural way so
that similar keys are not far away. The advantage is
that fewer nodes have to be accessed. The drawback
is the uneven distribution of the key-value pairs.
Prominent users
• @WalmartLabs[26] (previously Kosmix) uses Cassandra with SSD
• Amadeus IT Group uses Cassandra for some of their
back-end systems.
• Cloudkick uses Cassandra to store the server metrics
of their users.[31]
• Constant Contact uses Cassandra in their email and
social media marketing applications.[32] Over 200
nodes are deployed.
• Digg, a large social news website, announced on
Sep 9th, 2009 that it is rolling out its use of
Cassandra[33] and confirmed this on March 8,
2010.[34] TechCrunch has since linked Cassandra to
Digg v4 reliability criticisms and recent company
struggles.[35] Lead engineers at Digg later rebuked
these criticisms as red herring and blamed a lack of
load testing.[36]
• Facebook used Cassandra to power Inbox Search,
with over 200 nodes deployed.[37] This was abandoned in late 2010 when they built Facebook Messaging platform on HBase as they “found Cassandra’s eventual consistency model to be a difficult
• Formspring uses Cassandra to count responses, as
well as store Social Graph data (followers, following,
blockers, blocking) for 26 Million accounts with 10
million responses a day[39]
• IBM has done research in building a scalable email
system based on Cassandra.[40]
• uses Cassandra to record user activity
logs and topics for their Q&A website[41][42]
• Netflix uses Cassandra as their back-end database
for their streaming services[43][44]
• Nutanix appliances use Cassandra to store metadata
and stats.[45]
• Ooyala built a scalable, flexible, real-time analytics
engine using Cassandra[46]
• Openwave uses Cassandra as a distributed database
and as a distributed storage mechanism for their next
generation messaging platform[47]
• Apple uses 100,000 Cassandra nodes, as revealed at
Cassandra Summit San Francisco 2015,[27] although
it has not elaborated for which products, services or
• OpenX is running over 130 nodes on Cassandra
for their OpenX Enterprise product to store and
replicate advertisements and targeting data for ad
• AppScale uses Cassandra as a back-end for Google
App Engine applications[28]
• Plaxo has “reviewed 3 billion contacts in [their]
database, compared them with publicly available
data sources, and identified approximately 600 million unique people with contact info.”[49]
• CERN used Cassandra-based prototype for its
ATLAS experiment to archive the online DAQ system’s monitoring information[29]
• Cisco's WebEx uses Cassandra to store user feed
and activity in near real time.[30]
• PostRank used Cassandra as their backend
• Rackspace is known to use Cassandra internally.[51]
• Reddit switched to Cassandra from memcacheDB Commercial companies
on March 12, 2010[52] and experienced some prob• DataStax
lems in May due to insufficient nodes in their
• Impetus Technologies
• RockYou uses Cassandra to record every single click
for 50 million Monthly Active Users in real-time for
their online games[54]
• SoundCloud uses Cassandra to store the dashboard
of their users[55]
• Talentica Software uses Cassandra as a back-end
for Analytics Application with Cassandra cluster of
30 nodes and inserting around 200GB data on daily
• Tibbo Systems uses Cassandra as configuration and
event storage for AggreGate Platform.
• Twitter announced it was planning to move entirely
from MySQL to Cassandra,[57][58] though soon after retracted this, keeping Tweets in MySQL while
using Cassandra for analytics.[59]
• Instaclustr
• Apache Accumulo—Secure Apache Hadoop based
distributed database.
• Aerospike
• Berkeley DB
• Druid (open-source data store)
• Apache HBase—Apache Hadoop based distributed
database. Very similar to BigTable
• HyperDex
• Hypertable—Apache Hadoop based distributed
database. Very similar to BigTable
• Urban Airship uses Cassandra with the mobile service hosting for over 160 million application installs
across 80 million unique devices[60]
• MongoDB
• Wikimedia uses Cassandra as backend storage for
its public-facing REST Content API.[61]
• ScyllaDB
• Riak
• Zoho uses Cassandra for generating the inbox pre- 9.4.8
view in their Zoho#Zoho Mail service
Facebook moved off its pre-Apache Cassandra deployment in late 2010 when they replaced Inbox Search with
the Facebook Messaging platform.[38] In 2012, Facebook
began using Apache Cassandra in its Instagram unit.[62]
Cassandra is the most popular wide column store,[63] and
in September 2014 surpassed Sybase to become the 9th
most popular database, close behind Microsoft Access
and SQLite.[64]
See also
Academic background
• BigTable - Original distributed database by Google
• Distributed database
• Distributed hash table (DHT)
• Dynamo (storage system) - Cassandra borrows
many elements from Dynamo
[1] Casares, Joaquin (2012-11-05). “Multi-datacenter Replication in Cassandra”. DataStax. Retrieved 2013-07-25.
Cassandra’s innate datacenter concepts are important as
they allow multiple workloads to be run across multiple
[2] Rabl, Tilmann; Sadoghi, Mohammad; Jacobsen, HansArno; Villamor, Sergio Gomez-; Mulero -, Victor
Muntes; Mankovskii, Serge (2012-08-27). “Solving Big
Data Challenges for Enterprise Application Performance
Management” (PDF). VLDB. Retrieved 2013-07-25. In
terms of scalability, there is a clear winner throughout our
experiments. Cassandra achieves the highest throughput
for the maximum number of nodes in all experiments...
this comes at the price of high write and read latencies
[3] Hamilton, James (July 12, 2008). “Facebook Releases
Cassandra as Open Source”. Retrieved 2009-06-04.
[4] “Is this the new hotness now?". 200903-02. Archived from the original on 25 April 2010. Retrieved 2010-03-29.
[5] “Cassandra is an Apache top level project”. 2010-02-18. Archived from the original on
28 March 2010. Retrieved 2010-03-29.
[7] The Apache Software Foundation Announces Apache
Cassandra Release 0.6 : The Apache Software Foundation Blog
[8] The Apache Software Foundation Announces Apache
Cassandra 0.7 : The Apache Software Foundation Blog
[9] [Cassandra-user] [RELEASE] 0.8.0 - Grokbase
[10] Cassandra 1.0.0. Is Ready for the Enterprise
[11] The Apache Software Foundation Announces Apache
Cassandra™ v1.1 : The Apache Software Foundation
[12] “The Apache Software Foundation Announces Apache
Cassandra™ v1.2 : The Apache Software Foundation
Blog”. Retrieved 11 December 2014.
[13] Eric Evans. "[Cassandra-User] [RELEASE] Apache Cassandra 2.0.4”. Retrieved 11 December 2014.
[14] Sylvain Lebresne (10 September 2014).
SUCCESS] Release Apache Cassandra 2.1.0”. Retrieved 11 December 2014.
[15] “Deploying Cassandra across Multiple Data Centers”.
DataStax. Retrieved 11 December 2014.
assigned to nodes. Thus nodes will end up storing preponderances of keys (and the associated data) corresponding
to one column family or another. If as is likely column
families store differing quantities of data with their keys,
or store data accessed according to differing usage patterns, then some nodes will end up with disproportionately more data than others, or serving more “hot” data
than others.
[26] "@WalmartLabs”. Retrieved 11 December 2014.
[27] Luca Martinetti: Apple runs more than 100k [production]
Cassandra nodes. on Twitter
[28] “Datastores on Appscale”.
[29] “A Persistent Back-End for the ATLAS Online Information Service (P-BEAST)".
[30] “Re: Cassandra users survey”. 200911-21. Archived from the original on 17 April 2010. Retrieved 2010-03-29.
[31] 4 Months with Cassandra, a love story |Cloudkick, manage servers better
[16] DataStax (2013-01-15). “About data consistency”. Retrieved 2013-07-25.
[32] Finley, Klint (2011-02-18). “This Week in Consolidation:
HP Buys Vertica, Constant Contact Buys Bantam Live and
More”. Read Write Enterprise.
[17] “Hadoop Support” article on Cassandra’s wiki
[33] Eure, Ian. “Looking to the future with Cassandra”.
[18] “DataStax C/C++ Driver for Apache Cassandra”. DataStax. Retrieved 15 December 2014.
[34] Quinn, John. “Saying Yes to NoSQL; Going Steady with
[35] Schonfeld, Erick. “As Digg Struggles, VP Of Engineering
Is Shown The Door”.
[20] Ellis, Jonathan (2012-02-15). “Schema in Cassandra
1.1”. DataStax. Retrieved 2013-07-25.
[21] Ellis, Jonathan (2010-12-03). “What’s new in Cassandra
0.7: Secondary indexes”. DataStax. Retrieved 2013-0725.
[22] Ellis, Jonathan (2012-03-02). “The Schema Management
Renaissance in Cassandra 1.1”. DataStax. Retrieved
[23] Lebresne, Sylvain (2012-08-05). “Coming in 1.2: Collections support in CQL3”. DataStax. Retrieved 201307-25.
[24] DataStax. “Apache Cassandra 0.7 Documentation - Column Families”. Apache Cassandra 0.7 Documentation.
Retrieved 29 October 2012.
[25] Williams, Dominic. “Cassandra: RandomPartitioner
vs OrderPreservingPartitioner”. Retrieved 2011-03-23. When building a
Cassandra cluster, the “key” question (sorry, that’s weak)
is whether to use the RandomPartitioner (RP), or the OrdengPartitioner (OPP). These control how your data is
distributed over your nodes. Once you have chosen your
partitioner, you cannot change without wiping your data,
so think carefully! The problem with OPP: If the distribution of keys used by individual column families is different, their sets of keys will not fall evenly across the ranges
[36] “Is Cassandra to Blame for Digg v4’s Failures?".
[37] “Niet compatibele browser”. Facebook. Retrieved 201003-29.
[38] Muthukkaruppan, Kannan. “The Underlying Technology
of Messages”.
[39] Cozzi, Martin (2011-08-31). “Cassandra at Formspring”.
[40] “BlueRunner: Building an Email Service in the Cloud”
(PDF). 2009-07-20. Retrieved 2010-03-29.
[41] “ powered by Apache Cassandra™" (PDF). Santa Clara, CA, USA: DataStax. 201204-10. Retrieved 2014-06-13.
[42] Watch Cassandra at |DataStax Episodes |Blip
[43] Cockcroft, Adrian (2011-07-11). “Migrating Netflix from
Datacenter Oracle to Global Cassandra”.
Retrieved 2014-06-13.
[44] Izrailevsky, Yury (2011-01-28). “NoSQL at Netflix”.
[45] “Nutanix Bible”.
[46] Ooyala (2010-05-18). “Designing a Scalable Database for
Online Video Analytics” (PDF). Mountain
View CA, USA. Retrieved 2014-06-14.
[47] Mainstay LLC (2013-11-11). “DataStax Case Study
of Openwave Messaging” (PDF). Santa
Clara, CA, USA: DataStax. Retrieved 2014-06-15.
[48] Ad Serving Technology - Advanced Optimization, Forecasting, & Targeting |OpenX
[49] Smalley, Preston (2011-03-20). “An important milestone
- and it’s only the beginning!".
[50] Grigorik, Ilya (2011-03-29). “Webpulp TV: Scaling PostRank with Ilya Grigorik”.
[51] “Hadoop and Cassandra (at Rackspace)".
2010-04-23. Retrieved 2011-09-01.
Stu Hood.
[52] david [ketralnis] (2010-03-12). “what’s new on reddit:
She who entangles men”. blog.reddit. Archived from the
original on 25 March 2010. Retrieved 2010-03-29.
[53] Posted by the reddit admins at (2010-05-11). “blog.reddit
-- what’s new on reddit: reddit’s May 2010 “State of the
Servers” report”. blog.reddit. Archived from the original
on 14 May 2010. Retrieved 2010-05-16.
[54] Pattishall, Dathan Vance (2011-03-23). “Cassandra is my
NoSQL Solution but”.
[55] “Cassandra at SoundCloud”.
[56] cite web|url=[]
[57] Popescu, Alex. “Cassandra @ Twitter: An Interview with
Ryan King”. myNoSQL. Archived from the original on 1
March 2010. Retrieved 2010-03-29.
[58] Babcock, Charles. “Twitter Drops MySQL For Cassandra
- Cloud databases”. InformationWeek. Archived from the
original on 2 April 2010. Retrieved 2010-03-29.
[59] King, Ryan (2010-07-10). “Cassandra at Twitter Today”. San Francisco, CA, USA: Twitter. Retrieved 2014-06-20.
[60] Onnen, Erik. “From 100s to 100s of Millions”.
[61] Wicke, Gabriel. “Wikimedia REST content API is now
available in beta”.
[62] Rick Branson (2013-06-26). “Cassandra at Instagram”.
DataStax. Retrieved 2013-07-25.
[63] DB-Engines. “DB-Engines Ranking of Wide Column
9.4.10 External links
• Lakshman, Avinash (2008-08-25). “Cassandra - A
structured storage system on a P2P Network”. Engineering @ Facebook’s Notes. Retrieved 2014-0617.
• “The Apache Cassandra Project”. Forest Hill, MD,
USA: The Apache Software Foundation. Retrieved
• “Project Wiki”. Forest Hill, MD, USA: The Apache
Software Foundation. Retrieved 2014-06-17.
• Hewitt, Eben (2010-12-01). “Adopting Apache
Cassandra”. InfoQ, C4Media Inc. Retrieved 2014-06-17.
• Lakshman, Avinash; Malik, Prashant (2009-08-15).
“Cassandra - A Decentralized Structured Storage
System” (PDF). The authors are
from Facebook. Retrieved 2014-06-17.
• Ellis, Jonathan (2009-07-29). “What Every Developer Should Know About Database Scalability”. Retrieved 2014-06-17. From
the OSCON 2009 talk on RDBMS vs. Dynamo,
BigTable, and Cassandra.
• “Cassandra-RPM - Red Hat Package Manager
(RPM) build for the Apache Cassandra project”. Menlo Park, CA, USA: Google
Project Hosting. Retrieved 2014-06-17.
• Roth, Gregor (2012-10-14). “Cassandra by example - the path of read and write requests”. Retrieved 2014-06-17.
• Mansoor, Umer (2012-11-04). “A collection of
Cassandra tutorials”. Retrieved 2015-02-08.
• Bushik, Sergey (2012-10-22).
“A vendorindependent comparison of NoSQL databases: Cassandra, HBase, MongoDB, Riak”. NetworkWorld.
Framingham, MA, USA and Staines, Middlesex,
UK: IDG. Retrieved 2014-06-17.
9.5 Berkeley DB
[64] DB-Engines. “DB-Engines Ranking”.
Berkeley DB (BDB) is a software library that provides a
high-performance embedded database for key/value data.
Berkeley DB is written in C with API bindings for C++,
9.4.9 Bibliography
C#, PHP, Java, Perl, Python, Ruby, Tcl, Smalltalk, and
many other programming languages. BDB stores arbi• Hewitt, Eben (December 15, 2010). Cassandra: trary key/data pairs as byte arrays, and supports multiple
The Definitive Guide (1st ed.). O'Reilly Media. p. data items for a single key. Berkeley DB is not a relational
300. ISBN 978-1-4493-9041-9.
• Capriolo, Edward (July 15, 2011). Cassandra High BDB can support thousands of simultaneous threads of
Performance Cookbook (1st ed.). Packt Publishing. control or concurrent processes manipulating databases
p. 324. ISBN 1-84951-512-3.
as large as 256 terabytes,[2] on a wide variety of operating
systems including most Unix-like and Windows systems,
and real-time operating systems. “Berkeley DB” is also
used as the common brand name for three distinct products: Oracle Berkeley DB, Berkeley DB Java Edition, and
Berkeley DB XML. These three products all share a common ancestry and are currently under active development
at Oracle Corporation.
Berkeley DB originated at the University of California,
Berkeley as part of BSD, Berkeley’s version of the Unix
operating system. After 4.3BSD (1986), the BSD developers attempted to remove or replace all code originating in the original AT&T Unix from which BSD was
derived. In doing so, they needed to rewrite the Unix
database package.[3] A non-AT&T-copyrighted replacement, due to Seltzer and Yigit,[4] attempted to create a
disk hash table that performed better than any of the existing Dbm libraries. Berkeley DB itself was first released in 1991 and later included with 4.4BSD.[3] In 1996
Netscape requested that the authors of Berkeley DB improve and extend the library, then at version 1.86, to suit
Netscape’s requirements for an LDAP server[5] and for
use in the Netscape browser. That request led to the creation of Sleepycat Software. This company was acquired
by Oracle Corporation in February 2006, which continues to develop and sell Berkeley DB.
Since its initial release, Berkeley DB has gone through
various versions. Each major release cycle has introduced
a single new major feature generally layering on top of
the earlier features to add functionality to the product.
The 1.x releases focused on managing key/value data storage and are referred to as “Data Store” (DS). The 2.x releases added a locking system enabling concurrent access
to data. This is what is known as “Concurrent Data Store”
(CDS). The 3.x releases added a logging system for transactions and recovery, called “Transactional Data Store”
(TDS). The 4.x releases added the ability to replicate log
records and create a distributed highly available singlemaster multi-replica database. This is called the “High
Availability” (HA) feature set. Berkeley DB’s evolution
has sometimes led to minor API changes or log format
changes, but very rarely have database formats changed.
Berkeley DB HA supports online upgrades from one version to the next by maintaining the ability to read and apply the prior release’s log records.
cense, which is an OSI-approved open-source license as
well as an FSF-approved free software license.[8][9] The
product ships with complete source code, build script, test
suite, and documentation. The code quality and general
utility along with the licensing terms have led to its use in
a multitude of free and open-source software. Those who
do not wish to abide by the terms of the GNU AGPL, or
use an older version with the Sleepycat Public License,
have the option of purchasing another proprietary license
for redistribution from Oracle Corporation. This technique is called dual licensing.
Berkeley DB includes compatibility interfaces for some
historic Unix database libraries: dbm, ndbm and hsearch
(a System V library for creating in-memory hash tables).
9.5.2 Architecture
Berkeley DB has an architecture notably simpler than that
of other database systems like relational database management systems. For example, like SQLite, it does not
provide support for network access — programs access
the database using in-process API calls. Oracle added
support for SQL in 11g R2 release based on the popular
SQLite API by including a version of SQLite in Berkeley DB.[10] There is third party support for PL/SQL in
Berkeley DB via a commercial product named Metatranz
A program accessing the database is free to decide how
the data is to be stored in a record. Berkeley DB puts no
constraints on the record’s data. The record and its key
can both be up to four gigabytes long.
Despite having a simple architecture, Berkeley DB supports many advanced database features such as ACID
transactions, fine-grained locking, hot backups and
9.5.3 Editions
The name “Berkeley DB” is given to three different products:
1. Berkeley DB
2. Berkeley DB Java Edition
3. Berkeley DB XML
The FreeBSD and OpenBSD operating systems continue
to use Berkeley DB 1.8x for compatibility reasons;[6]
Linux-based operating systems commonly include sev- Each edition has separate database libraries, despite the
eral versions to accommodate for applications still using common branding. The first is the traditional Berkeolder interfaces/files.
ley DB, written in C. It contains several database imStarting with the 6.0/12c releases, all Berkeley DB prod- plementations, including a B-Tree and one built around
ucts are licensed under the GNU AGPL.[7] Up until then extendible hashing. It supports multiple language bindBerkeley DB was redistributed under the 4-clause BSD ings, including C/C++, Java (via JNI), C# .NET, Perl and
license (before version 2.0), and the Sleepycat Public Li- Python.
Berkeley DB Java Edition (JE) is a pure Java database
management library. Its design resembles that of Berkeley DB without replicating it exactly, and has a feature
set that includes many of those found in the traditional
Berkeley DB and others that are specific to the Java Edition. It has a log structured storage architecture, which
gives it different performance and concurrency characteristics. Three APIs are available—a Direct Persistence
Layer which is “Plain Old Java Objects” (POJO); one
which is based on the Java Collections Framework (an
object persistence approach); and one based on the traditional Berkeley DB API. The Berkeley DB Java Edition
High Availability option (Replication) is available. Note
that traditional Berkeley DB also supports a Java API, but
it does so via JNI and thus requires an installed native library.
The Berkeley DB XML database specializes in the storage of XML documents, supporting XQuery via XQilla.
It is implemented as an additional layer on top of (a legacy
version of) Berkeley DB and the Xerces library. DB
XML is written in C++ and supports multiple language
bindings, including C++, Java (via JNI), Perl and Python.
Programs that use Berkeley DB
Berkeley DB provides the underlying storage and retrieval
system of several LDAP servers, database systems, and
many other proprietary and free/open source applications. Notable software that use Berkeley DB for data
storage include:
• 389 Directory Server - An open-source LDAP
server from the Fedora Project.
• Bitcoin Core - The first implementation of the Bitcoin cryptocurrency.
• KDevelop – An IDE for Linux and other Unix-like
operating systems
• Movable Type (until version 4.0) – A proprietary
weblog publishing system developed by Californiabased Six Apart
• memcachedb - A persistence-enabled variant of
• MySQL database system – Prior to v5.1, MySQL
included a BDB data storage backend.
• OpenLDAP – A free/open source implementation of the Lightweight Directory Access Protocol
• Oracle NoSQL - A NoSQL distributed key-value
• Oracle Retail Predictive Application Server (RPAS)
- RPAS (since 12.x?) uses Berkeley DB as the underlying persistence layer for the MOLAP engine
used in several Oracle Retail Planning and Supply
Chain products. (Berkeley DB replaced the previous Accumate/Acumen persistence layer used since
the original development of the RPAS product by
Neil Thall Associates, which was no longer supported by its final owner, Lucent and no longer sufficiently scalable).
• Postfix – A fast, secure, easy-to-administer MTA for
Linux/Unix systems
• Parallel Virtual File System (PVFS) – A parallel file
system for HPC clusters.[12]
• Red Dwarf - A server framework originally developed by Sun, now open sourced, commonly used for
game development.
• Bogofilter – A free/open source spam filter that saves
its wordlists using Berkeley DB.
• RPM – The RPM Package Manager uses Berkeley
DB to retain its internal database of packages installed on a system
• Carbonado – An open source relational database access layer.
• Sendmail - A popular MTA for Linux/Unix systems
• Citadel – A free/open source groupware platform
that keeps all of its data stores, including the message base, in Berkeley DB.
• Subversion – A version control system designed
specifically to replace CVS
• Cyrus IMAP Server – A free/open source IMAP and
POP3 server, developed by Carnegie Mellon University
• Evolution - A free/open source mail client; contacts
are stored in addressbook.db using Berkeley DB
• Spamassassin – An anti-spam application
• Sun Grid Engine – A free and open source distributed resource management system.
9.5.5 Licensing
Oracle Corporation makes versions 2.0 and higher of
• GlusterFS - Prior to v3.4, GlusterFS included a BDB Berkeley DB available under a dual license. The Sleepycat license is a 2-clause BSD license with an additional
storage translator.
copyleft clause similar to the GNU GPL version 2’s Sec• Jabberd2 – An Extensible Messaging and Presence tion 3, requiring source code of an application using
Protocol server
Berkeley DB to be made available for a nominal fee.
As of Berkeley DB release 6.0, the Oracle Corporation
has relicensed Berkeley DB under the GNU AGPL v3.[14]
As of July 2011, Oracle’s list price for non-copyleft
Berkeley DB licenses varies between 900 and 13,800
USD per processor.[15] Embedded usage within the
Oracle Retail Predictive Application Server (RPAS) does
not require an additional license payment.
Sleepycat License
[3] Olson, Michael A.; Bostic, Keith; Seltzer, Margo (1999).
“Berkeley DB” (PDF). Proc. FREENIX Track, USENIX
Annual Tech. Conf. Retrieved October 20, 2009.
[4] Seltzer, Margo; Yigit, Ozan (1991). “A New Hashing
Package for UNIX”. Proc. USENIX Winter Tech. Conf.
Retrieved October 20, 2009.
[5] Brunelli, Mark (March 28, 2005). “A Berkeley DB
primer”. Enterprise Linux News. Retrieved December 28,
[6] “db(3)". Retrieved April 12, 2009.
Sleepycat License (sometimes referred to as Berkeley
Database License or the Sleepycat Public License) is [7] [Berkeley DB Announce] Major Release: Berkeley DB
12gR1 ( Retrieved July 5, 2013.
an OSI-approved open source license used by Oracle Corporation for the open-source editions of Berkeley DB, [8] “The Sleepycat License”. Open Source Initiative. October
Berkeley DB Java Edition and Berkeley DB XML em31, 2006. Retrieved December 28, 2008.
bedded database products older than version 6.0. (Starting with version 6.0 the open-source editions are instead [9] “Licenses”. Free Software Foundation. December 10,
2008. Retrieved December 28, 2008.
licensed under the GNU AGPL v3.) The name of this
license is derived from the name of the company which [10] “Twitter / Gregory Burd: @humanications We didn't r ...”.
commercially sold the Berkeley DB products, Sleepycat
Software, which was acquired by Oracle in 2006. Or- [11] “Official Berkeley DB FAQ”. Oracle Corporation. Retrieved March 30, 2010. Does Berkeley DB support
acle continued to use the name “Sleepycat License” for
Berkeley DB, despite not using the term “Sleepycat” in
any other documentation until it changed to GNU AGPL
[12] RCE 35: PVFS Parallel Virtual FileSystem
with version 6.
According to the Free Software Foundation,[16] it quali- [13] “Open Source License for Berkeley DB”. Oracle Corporation. For a license to use the Berkeley DB software unfies as a free software license, and is compatible with the
der conditions other than those described here, or to purGPL.
chase support for this software, please contact [email protected]
The license is a strong form of copyleft because it mandates that redistributions in any form not only include the
[14] “Major Release: Berkeley DB 12gR1 (". June
source code of Berkeley DB, but also “any accompany10, 2013. Retrieved July 15, 2013.
ing software that uses the DB software”. It is possible
to circumvent this strict licensing policy through the pur- [15]
chase of a commercial software license from Oracle Corporation consisting of terms and conditions which are ne[16] Various Licenses and Comments about Them - Free Softgotiated at the time of sale. This is an example of dual
ware Foundation
The effect of the dual license creates financial exposure [17] Mike Olson (co-founder and CEO of Sleepycat Software and Cloudera), lecture to Stanford University enfor commercial organizations, since there is considerable
trepreneurship students, 2013.11.13
risk of becoming liable for payment of license fees to Oracle. Some people consider it to be a “sneaky” license.
Mike Olson, co-founder and CEO of Sleepycat Software 9.5.7 External links
and Cloudera, said that “This is good business if you can
get it, but your relationship with your customer begins
• Oracle Berkeley DB Site
based on a threat and that’s not a really healthy place to
• Berkeley DB Programmer’s Reference Guide
start out.”[17]
• Licensing pitfalls for Oracle Technology Products
• The Berkeley DB Book by Himanshu Yadava
[1] Berkeley DB Reference Guide: What is Berkeley DB not?. (2001-05-31). Retrieved on 201309-18.
• - Berkeley DB at Launchpad
[2] Berkeley DB Reference Guide: Database limits Retrieved on
• Oracle Berkeley DB Licensing Information
• Oracle Licensing Knowledge Net
• Text of the Sleepycat License (old)
9.6 Memcached
Memcached (pronunciation: mem-cash-dee) is a
general-purpose distributed memory caching system.
It is often used to speed up dynamic database-driven
websites by caching data and objects in RAM to reduce
the number of times an external data source (such as a
database or API) must be read.
which, by default, expose their service at port 11211.
Each client knows all servers; the servers do not communicate with each other. If a client wishes to set or read the
value corresponding to a certain key, the client’s library
first computes a hash of the key to determine which server
to use. Then it contacts that server. This gives a simple
form of sharding and scalable shared-nothing architecture across the servers. The server computes a second
hash of the key to determine where to store or read the
corresponding value.
Memcached is free and open-source software, licensed
under the Revised BSD license.[2] Memcached runs on
Unix-like operating systems (at least Linux and OS X) The servers keep the values in RAM; if a server runs out
and on Microsoft Windows. It depends on the libevent of RAM, it discards the oldest values. Therefore, clients
must treat Memcached as a transitory cache; they cannot
assume that data stored in Memcached is still there when
Memcached’s APIs provide a very large hash table dis- they need it. Other databases, such as MemcacheDB,
tributed across multiple machines. When the table is full, Couchbase Server, provide persistent storage while mainsubsequent inserts cause older data to be purged in least taining Memcached protocol compatibility.
recently used (LRU) order.[3][4] Applications using Memcached typically layer requests and additions into RAM If all client libraries use the same hashing algorithm to debefore falling back on a slower backing store, such as a termine servers, then clients can read each other’s cached
The size of this hash table is often very large. It is limited A typical deployment has several servers and many
to available memory across all the servers in the cluster of clients. However, it is possible to use Memcached on
servers in a data centre. Where high volume, wide audi- a single computer, acting simultaneously as client and
ence web publishing requires it, this may stretch to many server.
gigabytes. Memcached can be equally valuable for situations where either the number of requests for content is
high, or the cost of generating a particular piece of con- Security
tent is high.
Most deployments of Memcached are within trusted netMemcached was originally developed by Danga Inter- works where clients may freely connect to any server.
active for LiveJournal, but is now used by many other However, sometimes Memcached is deployed in unsystems, including MocoSpace,[5] YouTube,[6] Reddit,[7] trusted networks or where administrators want to exercise
Survata,[8] Zynga,[9] Facebook,[10][11][12] Orange,[13] control over the clients that are connecting. For this purTwitter,[14] Tumblr[15] and Wikipedia.[16] Engine Yard pose Memcached can be compiled with optional SASL
and Jelastic are using Memcached as part of their authentication support. The SASL support requires the
platform as a service technology stack[17][18] and Heroku binary protocol.
offers several Memcached services[19] as part of their
platform as a service. Google App Engine, AppScale, A presentation at BlackHat USA 2010 revealed that a
Microsoft Azure and Amazon Web Services also offer a number of large public websites had left Memcached
open to inspection, analysis, retrieval, and modification
Memcached service through an API.[20][21][22][23]
of data.[28]
Even within a trusted organisation, the flat trust model of
memcached may have security implications. For efficient
simplicity, all Memcached operations are treated equally.
Memcached was first developed by Brad Fitzpatrick for Clients with a valid need for access to low-security enhis website LiveJournal, on May 22, 2003.[24][25][26] It tries within the cache gain access to all entries within the
was originally written in Perl, then later rewritten in C cache, even when these are higher-security and that client
by Anatoly Vorobey, then employed by LiveJournal.[27]
has no justifiable need for them. If the cache key can be
either predicted, guessed or found by exhaustive searching, its cache entry may be retrieved.
Software architecture
The system uses a client–server architecture. The servers
maintain a key–value associative array; the clients populate this array and query it by key. Keys are up to 250
bytes long and values can be at most 1 megabyte in size.
Clients use client-side libraries to contact the servers
Some attempt to isolate setting and reading data may be
made in situations such as high volume web publishing.
A farm of outward-facing content servers have read access to memcached containing published pages or page
components, but no write access. Where new content is
published (and is not yet in memcached), a request is instead sent to content generation servers that are not pub-
lically accessible to create the content unit and add it to 9.6.4 See also
memcached. The content server then retries to retrieve it
• Aerospike
and serve it outwards.
Example code
• phpFastCache - Supported MemCached, MemCache, WinCache, APC and Files.
• Couchbase Server
Note that all functions described on this page are
pseudocode only. Memcached calls and programming languages may vary based on the API used.
Converting database or object creation queries to use
Memcached is simple. Typically, when using straight
database queries, example code would be as follows:
function get_foo(int userid) { data = db_select(“SELECT
* FROM users WHERE userid = ?", userid); return data;
After conversion to Memcached, the same call might look
like the following
• Redis
• Mnesia
• MemcacheDB
• MySQL - directly supports the Memcached API as
of version 5.6.[29]
• Oracle Coherence - directly supports the Memcached API as of version 12.1.3.[30]
• GigaSpaces XAP - support Memcached with high
availability, transaction support[31]
• Hazelcast
function get_foo(int userid) { /* first try the cache
*/ data = memcached_fetch(“userrow:" + userid); if
• Cassandra
(!data) { /* not found : request database */ data =
db_select(“SELECT * FROM users WHERE userid
= ?", userid); /* then store in cache until next get */ 9.6.5 References
memcached_add(“userrow:" + userid, data); } return
[1] “Release notes for Release 1.4.22”. Retrieved 2015-04data; }
The client would first check whether a Memcached value
with the unique key "userrow:userid" exists, where userid
is some number. If the result does not exist, it would select from the database as usual, and set the unique key
using the Memcached API add function call.
However, if only this API call were modified, the server
would end up fetching incorrect data following any
database update actions: the Memcached “view” of the
data would become out of date. Therefore, in addition
to creating an “add” call, an update call would also be
needed using the Memcached set function.
[2] “Memcached license”. GitHub. Retrieved 2014-06-27.
[3] “Memcached NewOverview”.
[4] “Memcached NewUserInternals”.
[5] MocoSpace Architecture - 3 Billion Mobile Page Views
a Month. High Scalability (2010-05-03). Retrieved on
[6] Cuong Do Cuong (Engineering manager at
YouTube/Google) (June 23, 2007). Seattle Conference on Scalability: YouTube Scalability (Online Video 26th minute). Seattle: Google Tech Talks.
function update_foo(int userid, string dbUpdat[7] Steve Huffman on Lessons Learned at Reddit
eString) { /* first update database */ result =
db_execute(dbUpdateString); if (result) { /* database [8]
update successful : fetch data to be stored in cache */
data = db_select(“SELECT * FROM users WHERE [9] How Zynga Survived FarmVille
userid = ?", userid); /* the previous line could also look [10] Facebook Developers Resources
like data = createDataFromDBString(dbUpdateString);
*/ /* then store in cache until next get */ mem- [11] Scaling Memcached at Facebook
cached_set(“userrow:" + userid, data); } }
[12] NSDI '13: Scaling Memcache at Facebook
[13] Orange Developers
This call would update the currently cached data to match
the new data in the database, assuming the database query [14] It’s Not Rocket Science, But It’s Our Work
succeeds. An alternative approach would be to invalidate
the cache with the Memcached delete function, so that [15] Engineer – Core Applications Group job at Tumblr
in New York, NY, powered by JobScore.
subsequent fetches result in a cache miss. Similar acRetrieved on 2013-09-18.
tion would need to be taken when database records were
deleted, to maintain either a correct or incomplete cache. [16] MediaWiki Memcached
[17] Engine Yard Technology Stack
[18] Jelastic Memcached System
[19] Heroku Memcached add-ons
[20] Using Memcache - Google App Engine - Google Code
[21] Memcached in AppScale
[22] About In-Role Cache for Windows Azure Cache. Retrieved on 2013-09-18.
[23] Amazon ElastiCache. Retrieved on
[24] changelog: livejournal.
(2003-05-22). Retrieved on 2013-09-18.
[25] brad’s life - weather, running, distributed cache daemon. (2003-05-22). Retrieved on 201309-18.
[26] lj_dev: memcached. (200305-27). Retrieved on 2013-09-18.
[27] lj_dev: memcached. (2003-0527). Retrieved on 2013-09-18.
[28] BlackHat Write-up: go-derper and mining memcaches
[29] “Speedy MySQL 5.6 takes aim at NoSQL, MariaDB.”
External links
• Official website
• Memcached wiki and faq
• PHP Memcached Manager with Tag Support
• membase
• Memcached and Ruby
• go-memcached - Memcached implementation in Go
9.7 BigTable
Bigtable is a compressed, high performance, and
proprietary data storage system built on Google File System, Chubby Lock Service, SSTable (log-structured storage like LevelDB) and a few other Google technologies. On May 6, 2015, a public version of Bigtable was
launched as Google Cloud Bigtable.[1] Bigtable also underlies Google Datastore,[2] which is available as a part
of the Google Cloud Platform.
9.7.1 History
Bigtable development began in 2004[3] and is now used by
a number of Google applications, such as web indexing,[4]
MapReduce, which is often used for generating and
modifying data stored in Bigtable,[5] Google Maps,[6]
Google Book Search, “My Search History”, Google
Earth,, Google Code hosting, YouTube,[7]
and Gmail.[8] Google’s reasons for developing its own
database include scalability and better control of performance characteristics.[9]
Google’s Spanner RDBMS is layered on an implementation of Bigtable with a Paxos group for two-phase commits to each table. Google F1 was built using Spanner to
replace an implementation based on MySQL.[10]
9.7.2 Design
Bigtable maps two arbitrary string values (row key and
column key) and timestamp (hence three-dimensional
mapping) into an associated arbitrary byte array. It is
not a relational database and can be better defined as
a sparse, distributed multi-dimensional sorted map.[11]:1
Bigtable is designed to scale into the petabyte range across
“hundreds or thousands of machines, and to make it easy
to add more machines [to] the system and automatically
start taking advantage of those resources without any
• QuickCached - Memcached server implementation
Each table has multiple dimensions (one of which is a
in Java
field for time, allowing for versioning and garbage col• nsmemcache - memcache client for AOL Server
lection). Tables are optimized for Google File System
(GFS) by being split into multiple tablets – segments of
• Memcached implementation on Windows 8/8.1
the table are split along a row chosen such that the tablet
will be ~200 megabytes in size. When sizes threaten to
grow beyond a specified limit, the tablets are compressed
Commercially supported distributions
using the algorithm BMDiff[13][14] and the Zippy com• Couchbase Server (formerly Membase) offers a pression algorithm[15] publicly known and open-sourced
Memcached “bucket type” (free for use, subscrip- as Snappy,[16] which is a less space-optimal variation of
tion support available)
LZ77 but more efficient in terms of computing time.
The locations in the GFS of tablets are recorded as
• GigaSpaces Java based Memcached (free commudatabase entries in multiple special tablets, which are
nity edition, fault tolerance)
called “META1” tablets. META1 tablets are found by
• Hazelcast Memcached clustered, elastic, fault- querying the single “META0” tablet, which typically retolerant, Java based Memcached (free for use, sub- sides on a server of its own since it is often queried by
scription support available)
clients as to the location of the “META1” tablet which
itself has the answer to the question of where the actual
data is located. Like GFS’s master server, the META0
server is not generally a bottleneck since the processor time and bandwidth necessary to discover and transmit META1 locations is minimal and clients aggressively
cache locations to minimize queries.
Other similar software
• Apache Accumulo — built on top of Hadoop,
ZooKeeper, and Thrift. Has cell-level access labels
and a server-side programming mechanism. Written
in Java.
• Apache Cassandra — brings together Dynamo's
fully distributed design and Bigtable’s data model.
Written in Java.
• Apache HBase — Provides Bigtable-like support on
the Hadoop Core.[17] Has cell-level access labels and
a server-side programming mechanism too. Written
in Java.
• Hypertable — Hypertable is designed to manage the
storage and processing of information on a large
cluster of commodity servers.[18] Written in C++.
• “KDI”, Bluefish, GitHub — Kosmix attempt to
make a Bigtable clone. Written in C++.
• LevelDB — Google’s embedded key/value store
that uses similar design concepts as the Bigtable
[4] Chang, Fay; Dean, Jeffrey; Ghemawat, Sanjay; Hsieh,
Wilson C; Wallach, Deborah A; Burrows, Michael ‘Mike’;
Chandra, Tushar; Fikes, Andrew; Gruber, Robert E
(2006), “Bigtable: A Distributed Storage System for
Structured Data”, Research (PDF), Google .
[5] Chang et al. 2006, p. 3: ‘Bigtable can be used with
MapReduce, a framework for running large-scale parallel computations developed at Google. We have written a
set of wrappers that allow a Bigtable to be used both as an
input source and as an output target for MapReduce jobs’
[6] Whitchcock, Andrew, Google’s BigTable, There are currently around 100 cells for services such as Print, Search
History, Maps, and Orkut.
[7] Cordes, Kyle (2007-07-12), YouTube Scalability (talk),
Their new solution for thumbnails is to use Google’s
BigTable, which provides high performance for a large
number of rows, fault tolerance, caching, etc. This is a
nice (and rare?) example of actual synergy in an acquisition..
[8] “How Entities and Indexes are Stored”, Google App Engine, Google Code.
[9] Chang et al. 2006, Conclusion: ‘We have described
Bigtable, a distributed system for storing structured data
at Google... Our users like the performance and high
availability provided by the Bigtable implementation, and
that they can scale the capacity of their clusters by simply adding more machines to the system as their resource
demands change over time... Finally, we have found that
there are significant advantages to building our own storage solution at Google. We have gotten a substantial
amount of flexibility from designing our own data model
for Bigtable.’
• Distributed data stores, an overview
[10] Shute, Jeffrey ‘Jeff’; Oancea, Mircea; Ellner, Stephan;
Handy, Benjamin ‘Ben’; Rollins, Eric; Samwel, Bart; Vingralek, Radek; Whipkey, Chad; Chen, Xin; Jegerlehner,
Beat; Littlefield, Kyle; Tong, Phoenix (2012), “Summary;
F1 — the Fault-Tolerant Distributed RDBMS Supporting
Google’s Ad Business”, Research (presentation), Sigmod:
Google, p. 19, We've moved a large and critical application suite from MySQL to F1.
• Dynamo (storage system)
[11] Chang et al. 2006.
• Column-oriented DBMS
[12] “Google File System and BigTable”, Radar (World Wide
Web log), Database War Stories (7), O’Reilly, May 2006.
See also
• Amazon SimpleDB
• Big data
• Hadoop
[3] Kumar, Aswini, Whitchcock, Andrew, ed., Google’s
BigTable, First an overview. BigTable has been in development since early 2004 and has been in active use for
about eight months (about February 2005)..
[13] “Google Bigtable, Compression, Zippy and BMDiff”.
2008-10-12. Archived from the original on 1 May 2013.
Retrieved 14 April 2015..
[14] McIlroy, Bentley. Data compression using long common
strings. DCC '99. IEEE..
[15] “Google’s Bigtable”, Outer court (Weblog), 2005-10-23.
[16] “Snappy”, Code (project), Google.
[17] “Background; HBase”, Hadoop Core (wiki), Apache.
[18] “About”, Hyper table.
[19] “Leveldb file layout and compactions”, Code, Google.
• Chang, Fay; Dean, Jeffrey; Ghemawat, Sanjay;
Hsieh, Wilson C; Wallach, Deborah A; Burrows,
Michael ‘Mike’; Chandra, Tushar; Fikes, Andrew;
Gruber, Robert E (2006), “Bigtable: A Distributed
Storage System for Structured Data”, Research
(PDF), Google .
External links
• BigTable: A Distributed Structured Storage System,
Washington. Video, Google.
• UWTV (video).
• Witchcock, Andrew, Google’s BigTable (notes
on the official presentation).
• Carr, David F (2006-07-06), “How Google Works”,
• “Is the Relational Database Doomed?", Read-write
Chapter 10
Text and image sources, contributors, and
10.1 Text
• Database Source: Contributors: Paul Drye, NathanBeach, Dreamyshade,
LC~enwiki, Robert Merkel, Zundark, The Anome, Stephen Gilbert, Sjc, Andre Engels, Chuckhoffmann, Fubar Obfusco, Ben-Zin~enwiki,
Maury Markowitz, Waveguy, Imran, Leandrod, Stevertigo, Edward, Ubiquity, Michael Hardy, JeffreyYasskin, Fuzzie, Pnm, Ixfd64,
TakuyaMurata, SebastianHelm, Pcb21, CesarB, MartinSpamer, ArnoLagrange, Ahoerstemeier, Haakon, Nanshu, Angela, Bogdangiusca,
Cyan, Poor Yorick, Mxn, Mulad, Feedmecereal, Jay, Greenrd, Wik, DJ Clayworth, Tpbradbury, E23~enwiki, Furrykef, Morwen, Sandman~enwiki, Finlay McWalter, Jni, Chuunen Baka, Robbot, Noldoaran, Sander123, Craig Stuntz, Chrism, Chris 73, Vespristiano, Chocolateboy, Netizen, Nurg, Romanm, Lowellian, Pingveno, Tualha, Rursus, Rothwellisretarded, Jondel, TittoAssini, Hadal, Vikreykja, Mushroom, HaeB, Pengo, SpellBott, Tobias Bergemann, Stirling Newberry, Psb777, Giftlite, Graeme Bartlett, SamB, Sarchand~enwiki, Arved,
Inter, Kenny sh, Levin, Peruvianllama, Everyking, Ciciban, Ssd, Niteowlneils, Namlemez, Mboverload, SWAdair, Bobblewik, Wmahan,
Gadfium, SarekOfVulcan, Quadell, Kevins, Antandrus, Beland, OverlordQ, Rdsmith4, APH, Troels Arvin, Gscshoyru, Ohka-, Sonett72,
Trevor MacInnis, Canterbury Tail, Bluemask, Zro, Grstain, Mike Rosoft, DanielCD, Shipmaster, EugeneZelenko, AnjaliSinha, KeyStroke,
Discospinster, Rich Farmbrough, Rhobite, Lovelac7, Pak21, C12H22O11, Andrewferrier, Mumonkan, Kzzl, Paul August, Edgarde, Djordjes, S.K., Elwikipedista~enwiki, CanisRufus, *drew, MBisanz, Karmafist, Kiand, Cpereyra, Tom, Causa sui, Chrax, PatrikR, Hurricane111,
Mike Schwartz, Smalljim, Wipe, John Vandenberg, Polluks, Ejrrjs, JeffTan, Nk, Franl, Alphax, Railgun, Sleske, Sam Korn, Nsaa, Mdd,
HasharBot~enwiki, Jumbuck, Storm Rider, Alansohn, Tablizer, Etxrge, Guy Harris, Arthena, Keenan Pepper, Ricky81682, Riana, AzaToth, Zippanova, Kocio, PaePae, Velella, Skybrian, Helixblue, Filx, Frankman, Danhash, Max Naylor, Harej, Mathewforyou, W mccall, Ringbang, Chirpy, Djsasso, Dan100, Brookie, Isfisk, Marasmusine, Simetrical, Reinoutr, Woohookitty, Mindmatrix, Camw, Arcann,
25or6to4, Decrease789, Mazca, Pol098, Commander Keane, Windsok, Ruud Koot, Tabletop, Bbatsell, KingsleyIdehen, DeirdreGerhardt,
GregorB, AnmaFinotera, Plrk, Crucis, Prashanthns, TrentonLipscomb, Turnstep, PeregrineAY, Dysepsion, Mandarax, Wulfila, MassGalactusUniversum, Graham87, Qwertyus, DePiep, Jclemens, Sjakkalle, Rjwilmsi, Koavf, DeadlyAssassin, Vary, Carbonite, GlenPeterson,
Feydey, Eric Burnett, Jb-adder, ElKevbo, The wub, Sango123, FlaBot, Doc glasgow, Latka, GnuDoyng, Jstaniek, RexNL, AndriuZ, Intgr,
Antrax, Ahunt, Imnotminkus, JonathanFreed, King of Hearts, Chobot, Visor, Phearlez, DVdm, Bkhouser, NSR, Cornellrockey, Rimonu,
YurikBot, Wavelength, Sceptre, JarrahTree, Phantomsteve, Michael Slone, Woseph, Fabartus, Toquinha, GLaDOS, SpuriousQ, RadioFan2 (usurped), Akamad, Stephenb, Rsrikanth05, Cryptic, Cpuwhiz11, BobStepno, Wimt, SamJohnston, RadioKirk, NawlinWiki, Wiki
alf, Jonathan Webley, Jaxl, Milo99, Welsh, Joel7687, SAE1962, Journalist, Nick, Aaron Brenneman, RayMetz100, Matticus78, Larsinio,
Mikeblas, Ezeu, Zwobot, Supten, Dbfirs, JMRyan, Bluerocket, LindaEllen, Samir, DeadEyeArrow, Werdna, User27091, Mugunth Kumar, SimonMorgan, Lod, Twelvethirteen, Deville, Theodolite, Zzuuzz, Mike Dillon, Closedmouth, Arthur Rubin, Fang Aili, Th1rt3en,
GraemeL, JoanneB, Alasdair, Echartre, JLaTondre, ArielGold, Stuhacking, Kungfuadam, Mhkay, Bernd in Japan, GrinBot~enwiki, DVD
R W, Jonearles, CIreland, Victor falk, Pillefj, SmackBot, Hydrogen Iodide, McGeddon, WikiuserNI, Unyoyega, Pgk, AnonUser, Davewild, AutumnSnow, Brick Thrower, Stifle, Jab843, PJM, Kslays, Edgar181, Lexo, David Fuchs, Siebren, Yamaguchi , Gilliam, Donama,
Ohnoitsjamie, Chaojoker, Chris the speller, TimBentley, MikeSy, Thumperward, Nafclark, Oli Filth, MalafayaBot, Silly rabbit, Robocoder,
Xx236, Deli nk, Jerome Charles Potts, Baa, Robth, DHN-bot~enwiki, Methnor, Colonies Chris, Darth Panda, Can't sleep, clown will
eat me, Frap, Chlewbot, Paul E Ester, Edivorce, Allan McInnes, Pax85, Mugaliens, Khoikhoi, COMPFUNK2, Soosed, Cybercobra,
Jwy, Jdlambert, Dreadstar, Insineratehymn, Hgilbert, BryanG, Ultraexactzz, RayGates, Daniel.Cardenas, Kukini, Kkailas, SashatoBot,
Krashlandon, Jasimab, Srikeit, Kuru, Jonwynne, Microchip08, Tazmaniacs, Gobonobo, PeeAeMKay, Sir Nicholas de Mimsy-Porpington,
Lguzenda, Tim Q. Wells, Minna Sora no Shita, Joffeloff, HeliXx, IronGargoyle, [email protected], MarkSutton, Slakr, Tasc, Beetstra, Noah Salzman,
Wikidrone, Babbling.Brook, Childzy, Optakeover, Waggers, Ryulong, ThumbFinger, DougBarry, Asyndeton, Dead3y3, Iridescent, Mrozlog, TwistOfCain, Paul Foxworthy, Igoldste, Benni39, Dwolt, DEddy, Courcelles, Linkspamremover, Navabromberger, Dkastner, Tawkerbot2, Flubeca, LessHeard vanU, Megatronium, FatalError, JForget, Comps, VoxLuna, Spdegabrielle, Thatperson, Ahy1, CmdrObot, Ale
jrb, Ericlaw02, Iced Kola, KyraVixen, Kushal one, GHe, Constructive, Dgw, Argon233, FlyingToaster, Moreschi, Sewebster, Simeon,
Joshnpowell, Ubiq, Cantras, Mato, Gogo Dodo, Parzi, Chasingsol, Pascal.Tesson, Dancter, SymlynX, Tawkerbot4, Shirulashem, DumbBOT, Chrislk02, Alaibot, IComputerSaysNo, SpK, Omicronpersei8, UberScienceNerd, Cavanagh, Click23, Mattisse, Thijs!bot, Epbr123,
Qwyrxian, HappyInGeneral, Andyjsmith, CynicalMe, Mojo Hand, Philippe, Eric3K, Peashy, Maxferrario, Mentifisto, AntiVandalBot,
Majorly, Luna Santin, Widefox, Seaphoto, Turlo Lomon, MrNoblet, EarthPerson, Kbthompson, Credema, Spartaz, Lfstevens, Deadbeef,
JAnDbot, Eric Bekins, MER-C, BlindEagle, The Transhumanist, Blood Red Sandman, RIH-V, Andonic, PhilKnight, Saiken79, LittleOldMe, Jdrumgoole, Magioladitis, Karlhahn, Bongwarrior, VoABot II, Hasek is the best, JamesBWatson, Think outside the box, Lucyin,
Twsx, WODUP, Cic, Jvhertum, Bubba hotep, Culverin, Danieljamesscott, Adrian J. Hunter, 28421u2232nfenfcenc, Stdazi, Wwmbes, Cpl
Syx, Kunaldeo, Kozmando, Chris G, DerHexer, JaGa, Ahodgkinson, Oroso, Leaderofearth, MartinBot, Ironman5247, Arjun01, NAHID,
Poeloq, CableCat, Rettetast, R'n'B, NetManage, Tgeairn, J.delanoy, Pharaoh of the Wizards, Trusilver, Rohitj.iitk, Bogey97, Ayecee, Uncle
Dick, Maurice Carbonaro, Jesant13, Ginsengbomb, Darth Mike, Gzkn, Bcartolo, BrokenSphere, Katalaveno, Afluegel, Chriswiki, DamnRandall, Girl2k, NewEnglandYankee, SJP, Gregfitzy, Kraftlos, Madth3, Madhava 1947, Jackacon, Juliancolton, Cometstyles, Ryager,
Raspalchima, Seanust 1, Lamp90, Bonadea, Aditya gopal3, Pdcook, Ja 62, TheNewPhobia, DigitalEnthusiast, Squids and Chips, CardinalDan, Ryanslater, Ryanslater2, Siteobserver, Lights, VolkovBot, Amaraiel, Thedjatclubrock, Alain Amiouni, Indubitably, JustinHagstrom,
WOSlinker, Barneca, N25696, Erikrj, Philip Trueman, Lingwitt, TXiKiBoT, Wiki tiki tr, Moogwrench, Vipinhari, Technopat, Caster23,
GDonato, ScMeGr, Olinga, Ann Stouter, Anonymous Dissident, Cyberjoac, Qxz, Gozzy345, Lradrama, Sintaku, Clarince63, Twebby,
BwDraco, Jackfork, LeaveSleaves, Wya 7890, Mannafredo, Amd628, Zhenqinli, Hankhuck, Gwizard, Synthebot, Kingius, Bblank, Why
Not A Duck, Atkinsdc, Pjoef, Aepanico, Logan, HybridBoy, Thehulkmonster, D. Recorder, Calutuigor, SieBot, Fooker69, Calliopejen1,
Praba tuty, Kimera Kat, Jauerback, LeeHam2007, Caltas, Eagleal, Triwbe, Yintan, TalkyLemon, Keilana, Bentogoa, Flyer22 Reborn,
Radon210, Oda Mari, JCLately, Jojalozzo, Hxhbot, Le Pied-bot~enwiki, Sucker666, Theory of deadman, KoshVorlon, 10285658sdsaa, Mkeranat, Fratrep, Macy, ChorizoLasagna, Autumn Wind, Maxime.Debosschere, Spazure, Paulinho28, Vanished User 8902317830,
G12kid, Pinkadelica, Treekids, Denisarona, Escape Orbit, Levin Carsten, Kanonkas, VanishedUser sdu9aya9fs787sads, Explicit, Beeblebrox, ClueBot, Frsparrow, Phoenix-wiki, Hugsandy, Strongsauce, Avenged Eightfold, The Thing That Should Not Be, Buzzimu, Jan1nad,
Poterxu, Supertouch, Unbuttered Parsnip, Garyzx, Zipircik, SuperHamster, Boing! said Zebedee, Doddsy1993, Niceguyedc, Sam Barsoom, Blanchardb, Dylan620, Mikey180791, Puchiko, Mspraveen, Vivacewwxu~enwiki, Veryprettyfish, Robert Skyhawk, Drumroll99,
Excirial, Pumpmeup, M4gnum0n, Dbates1999, LaosLos, Northernhenge, Eeekster, Tyler, Odavy, Cenarium, Lunchscale, Peter.C, Jotterbot, MECiAf., Huntthetroll, Tictacsir, Thehelpfulone, Inspector 34, Thingg, Aitias, DerBorg, Subash.chandran007, Versus22, Burner0718,
Johnuniq, SoxBot III, Apparition11, DumZiBoT, Jmanigold, XLinkBot, EastTN, Gsallis, PrePress, Avoided, Pee Tern, Galzigler, Noctibus,
Qwertykris, Dsimic, Osarius, HexaChord, Rakeki, Addbot, Proofreader77, Pyfan, Willking1979, Some jerk on the Internet, Betterusername, Non-dropframe, Captain-tucker, Ngpd, Fgnievinski, Fieldday-sunday, JephapE, Xhelllox, Vishnava, CanadianLinuxUser, Fluffernutter, Cevalsi, Cambalachero, CarsracBot, DFS454, Glane23, FiriBot, SDSWIKI, Roux, Favonian, Doniago, Exor674, AtheWeatherman,
Jasper Deng, Hotstaff, Evildeathmath, Tide rolls, Nicoosuna, Kivar2, Matěj Grabovský, Dart88, Gail, David0811, Duyanfang, Jarble,
Arbitrarily0, LuK3, Informatwr, Ben Ben, Luckas-bot, Yobot, Sudarevic, 2D, OrgasGirl, Bunnyhop11, Fraggle81, Gishac, MarcoAurelio, Pvjohnson, Nallimbot, SwisterTwister, Srdju001, Peter Flass, Bbb23, N1RK4UDSK714, AnomieBOT, AmritasyaPutra, Rubinbot,
Sonia, Jim1138, JackieBot, Piano non troppo, Kingpin13, Ulric1313, Imfargo, Flewis, Bluerasberry, Materialscientist, Kimsey0, Citation bot, OllieFury, BlurTento, Clark89, Darthvader023, Xqbot, Anders Torlind, Kimberly ayoma, Sythy2, Llyntegid, Addihockey10,
Capricorn42, Bcontins, 4twenty42o, Craftyminion, Grim23, Yossman007, Preet91119, Tonydent, GrouchoBot, Call me Bubba, Kekekecakes, Bjcubsfan, Earlypsychosis, Prunesqualer, Crashdoom, Amaury, Doulos Christos, Sophus Bie, The Wiki Octopus, IElonex!, Shadowjams, М И Ф, Oho1, Dougofborg, Al Wiseman, Chtuw, Captain-n00dle, Gonfus, Prari, FrescoBot, Sock, Riverraisin, Fortdj33, Blackguard SF, Dogposter, Mark Renier, StaticVision, HJ Mitchell, Sae1962, Wifione, Weetoddid, ZenerV, Kwiki, Javert, ZooPro, Winterst,
Shadowseas, Pinethicket, I dream of horses, HRoestBot, Grsmca, LittleWink, 10metreh, Supreme Deliciousness, Hamtechperson, Sissi’s
bd, 28nebraska, Jschnur, Xfact, RedBot, Btilm, MastiBot, Rotanagol, Bharath357, Σ, 05winsjp, Psaajid, Meaghan, Abhikumar1995,
Jandalhandler, Refactored, FoxBot, TobeBot, Mercy11, ‫کاشف عقیل‬, KotetsuKat, ItsZippy, Lotje, Callanecc, Writeread82, Vrenator,
Reidh21234, Reaper Eternal, Luizfsc, TheGrimReaper NS, Xin0427, Suffusion of Yellow, SnoFox, BluCreator, Colindolly, TheMesquito,
Minimac, Thinktdub, Heysim0n, RazorXX8, DARTH SIDIOUS 2, Lingliu07, KILLERKEA23, Onel5969, Leonnicholls07, Mean as custard, Helloher, ArwinJ, Kvasilev, Regancy42, FetchcommsAWB, Timbits82, Aj.robin, Salvio giuliano, Skamecrazy123, Rollins83, EmausBot, John of Reading, FFGeyer, Armen1304, Heymid, ScottyBerg, Lflores92201, Beta M, Dewritech, GoingBatty, RA0808, RenamedUser01302013, Itafran2010, Knbanker, Winner 42, Carbo1200, Wikipelli, K6ka, Sheeana, Ceyjan, Serketan, AsceticRose, Anirudh
Emani, Tudorol, Komal.Ar, Pete1248, Savh, Ravinjit, Joshwa1234567890, Fæ, NicatronTg, M.badnjki, Alpha Quadrant (alt), Tuhl,
Makecat, Ocaasi, OnePt618, Tolly4bolly, W163, TyA, L Kensington, Mayur, Donner60, Mentibot, MainFrame, RockMagnetist, Nz101,
Matthewrbowker, Peter Karlsen, GregWPhoto, GrayFullbuster, Rishu arora11, DASHBotAV, Kellyk99, 28bot, Rocketrod1960, Diamondland, ClueBot NG, SpikeTorontoRCP, Mechanical digger, Jack Greenmaven, MelbourneStar, Satellizer, Dancayta, Chester Markel,
Name Omitted, Bwhynot14, Millermk, Theimmaculatechemist, Lsschwar, Bowlderizer, Zhoravdb, Widr, Danim, Ugebgroup8, CasualVisitor, Vibhijain, Franky21, Jk2q3jrklse, Cammo33, Oddbodz, Lbausalop, Cambapp, Strike Eagle, Calabe1992, Doorknob747, Lowercase sigmabot, BG19bot, Freebiekr, MilerWhite, Machdohvah, Tomatronster, Northamerica1000, Wiki13, MusikAnimal, Frze, Dan653,
AwamerT, Allecher, Mark Arsten, Somchai1029, Vincent Liu, Compfreak7, 110808028 amol, Altaïr, Foxfax555, Rj Haseeb, Alzpp,
Bfugett, Jbrune, Thomasryno, Afree10, Glacialfox, Rowan Adams, Admfirepanther, Era7bd, Soumark, Maxmarengo, Manikandan 2030,
Branzman, Melucky2getu, Fylbecatulous, Plavozont, Carliitaeliza, IkamusumeFan, Several Pending, Pratyya Ghosh, Zhaofeng Li, Mrt3366,
VNeumann, ChrisGualtieri, Christophe.billiottet, Lovefamosos, Maty18, Mediran, Khazar2, Deathlasersonline, Saturdayswiki,
, Codename Lisa, Mukherjeeassociates, Cerabot~enwiki, Malvikiran, Cheolsoo, R3miixasim, Pebau.grandauer, TwoTwoHello, Lugia2453,
Frosty, SFK2, Graphium, Rafaelschp, 069952497a, Reatlas, Phamnhatkhanh, Epicgenius, P2Peter, Acetotyce, Rockonomics, Eyesnore,
Moazzam chand, JamesMoose, Jabby11, EvergreenFir, Menublogger, Backendgaming, PappaAvMin, Mike99999, Gburd, Babitaarora,
MJunkCat, Boli1107, JJdaboss, Ray Lightyear, BentlijDB, Hshoemark, Melody Lavender, Ginsuloft, D Eaketts, Eddiecarter1, Eddiejcarter, Mwaci11, Gajurahman, Manul, IrfanSha, AddWittyNameHere, Dkwebsub, JunWan, WPGA2345, Verajohne, Phinicle, Title302,
JaconaFrere, ElijahLloyd97, Suelru, 7Sidz, Monkbot, JewishMonser69, Rajat Kant Singh, Davidcopperfield123, Sunrocket89, Nomonomnom, Samster0708, Krushna124, Cabral88, MisteArndon, KizzyCode, Uoy ylgu dratsab, Hillysilly, FSahar, Thedinesh4u, Boybudz321,
Jesseminisis, ChamithN, Crystallizedcarbon, Eurodyne, JensLechtenboerger, Papapasan, Is8ac, Torvolt, Rgeurts, DiscantX, MaurolepisDreki, Top The Ball, Godfarther48, Jack0898, Rob12467, Gfsfg, Uthinkurspecial, Asdafsd, KasparBot, SaltySloth, Jrgreene2, Timoutiwin,
Smedley Rhyse-Frockmorton, Communal t, BadSprad, AldrinRemoto, Vinurarulz, Yeshii 909, Cani talk and Anonymous: 2088
• Schema migration Source: Contributors: Wtmitchell, Malcolma,
Racklever, Hu12, Paling Alchemist, AnomieBOT, Digulla, Billegge, LilHelpa, Diroussel, Snotbot, NietzscheSpeaks, Wokspoon, Andifalk, Axel.fontaine, Ebr.wolff, Dvdtknsn, JonRou, Bread2000, Gregoriomelo and Anonymous: 17
• Star schema Source: Contributors: Jketola, Random832, Jay, 1984, Remy
B, Pne, Dfrankow, Beland, Asqueella, KeyStroke, Appi~enwiki, .:Ajvol:., Gothick, Diego Moya, Andrewpmk, GregorB, DePiep,
Birger~enwiki, Chobot, YurikBot, Chrissi~enwiki, Od Mishehu, Vald, Mselway, Chronodm, Gilliam, Bluebot, Cralize, OrphanBot, RayGates, Michael miceli, Budyhead, JHunterJ, Bertport, Thesuperav, SqlPac, CWY2190, NishithSingh, Electrum, Kablammo, Armchairlinguist, Mwarren us, Littldo, Falcor84, Raymondwinn, Panfakes, Flyer22 Reborn, ClueBot, AndrewMWebster, Aitias, Addbot, Elsendero,
Luckas-bot, Yobot, Fraggle81, JackieBot, Materialscientist, Mark Renier, D'ohBot, Crysb, Gahooa, EmausBot, Txnate, ClueBot NG,
Gilderien, Ozancan, Mrityu411989, Sharad.sangle, Helpful Pixie Bot, BG19bot, Walk&check, Sitoiganap, Anubhab91, Jobin RV, Millertimebjm, Nrahimian, ChrisGualtieri, Abergquist, Surendra.konathala, Ginsuloft, Amattas, Andrew not the saint and Anonymous: 121
• CAP Source: Contributors: Gtrmp, RHaworth, BD2412, Wavelength, N35w101,
10.1. TEXT
Capmo, Clarityfiend, Connermcd, HelenOnline, Krassotkin, Brycehughes, ClueBot NG, Mark Arsten, Oranjblud, Gnasby, Alexwho314
and Anonymous: 6
• Eventual consistency Source: Contributors: The Anome, Alaric,
Charles Matthews, Finlay McWalter, Ruinia, Rich Farmbrough, Dodiad, Intgr, Emersoni, Fang Aili, Mshiltonj, SmackBot, Frap, JonHarder, Kinotgell, Gpierre, Gregbard, Cydebot, Kovan, Momo54, Duncan.Hull, Gdupont, Siskus, Morninj, Mild Bill Hiccup, M4gnum0n,
DumZiBoT, Addbot, Twimoki, Yobot, AnomieBOT, Materialscientist,, Theclapp, DataWraith, PeaceLoveHarmony, Erik9bot,
Skcpublic, Sae1962, Winterst, RedBot, GoingBatty, Mspreitz, ZéroBot, Richnice, Rob7139, ClueBot NG, BG19bot, SAuhsoj, Andrew
Helwer, APerson, Dexbot, Datamaniac, Pbailis and Anonymous: 23
• Object-relational impedance mismatch Source:
680987835 Contributors: Leandrod, GCarty, Rbraunwa, Morven, Topbanana, Jeffq, Craig Stuntz, Rursus, Ambarish, Bkonrad,
Esap, Jpp, SarekOfVulcan, Rich Farmbrough, Mike Schwartz, Mojo~enwiki, Pearle, Merenta, Tablizer, Diego Moya, Ruud Koot,
Triddle, Msiddalingaiah, Rjwilmsi, JubalHarshaw, MarSch, Salix alba, Dmccreary, Hairy Dude, Allister MacLeod, Big Brother 1984,
EngineerScotty, Grafen, ZacBowling, SAE1962, Ospalh, Scope creep, BazookaJoe, Fram, Draicone, Erik Postma, SmackBot, Brick
Thrower, Chris the speller, Thumperward, Jerome Charles Potts, Colonies Chris, Frap, Cybercobra, Warren, Zsvedic, Wickethewok,
Larrymcp, Hu12, CmdrObot, Jiminez, Arnonf, Pingku, Underpants, PKT, Towopedia, Ideogram, Mentifisto, Magioladitis, Dbasch,
Joshua Davis, Rustyfence, STBot, J.delanoy, Cantonnier, Q Chris, Andy Dingley, Prakash Nadkarni, Creative1985, M4gnum0n, Aprock,
Addbot, Sbhug1, N8allan, Agomulka, AnomieBOT, Roux-HG, Metafax1, Mark Renier, OldTownIT, GoingBatty, Rdmil, ClueBot NG,
Shaddim, Widr, Kcragin, Danim, MusikAnimal, Stelpa, Thesquaregroot, BattyBot, Fraulein451, Lesser Cartographies, Alexadamson and
Anonymous: 83
• Object database Source: Contributors: Vtan, Hari, Ben-Zin~enwiki,
Maury Markowitz, Leandrod, Stevertigo, W~enwiki, Modster, Kku, Pcb21, CesarB, Hofoen, Ronz, Rednblu, Furrykef, Robbot, Noldoaran,
BenFrantzDale, Beardo, Mckaysalisbury, Gadfium, Beland, SimonArlott, Sam Hocevar, Indolering, Usrnme h8er, Klemen Kocjancic,
Tordek ar, Pavel Vozenilek, Elwikipedista~enwiki, Enric Naval, Ejrrjs, Mdd, Tablizer, Zippanova, Rickyp, SteinbDJ, Voxadam, Forderud,
Karnesky, Mindmatrix, Dandv, Timosa, Ruud Koot, JIP, Icey, Jivecat, Dmccreary, FlaBot, SchuminWeb, Margosbot~enwiki, Intgr, Bgwhite, YurikBot, Foxxygirltamara, Hydrargyrum, ENeville, Oberst, SAE1962, Larsinio, Voidxor, Grafikm fr, BOT-Superzerocool, Ott2,
Sandstein, BSTRhino, Talyian, Wainstead, Mhkay, Mlibby, Dybdahl, SmackBot, Pintman, Kellen, Reedy, OEP, AutumnSnow, Commander
Keane bot, Bluebot, Nkaku, MyNameIsVlad, RProgrammer, Turbothy, Soumyasch, Lguzenda, IronGargoyle, RichardF, DougBarry, Hu12,
Britannica~enwiki, Tawkerbot2, DBooth, FatalError, LGuzenda, Ervinn, Shreyasjoshis, Singerboi22, Yaris678, Corpx, Sestoft, Torc2,
Charwing, Nick Number, Mvjs, Spencer, MER-C, Cameltrader, Jbom1, Magioladitis, Hroðulf, Soulbot, 28421u2232nfenfcenc, Gwern,
Wixardy, J.delanoy, Saifali1, Kozka, Tagus, VolkovBot, Rei-bot, WikipedianYknOK, Dawn Bard, JCLately, SmallRepair, Edlich, PsyberS,
VanishedUser sdu9aya9fs787sads, Dinojc, ClueBot, The Thing That Should Not Be, EoGuy, Alexbot, Eeekster, Sun Creator, Lucpeuvrier~enwiki, EastTN, HarlandQPitt, Zeliboba7, Addbot, Download, Kngspook, Lightbot, Jackelfive, Yobot, Bunnyhop11, Fraggle81,
Waynenilsen, AnomieBOT, Materialscientist, Xqbot, Mika au, Addbc, Garyaj, Pwwamic, FrescoBot, Sulfsby, Mark Renier, Bablind, ITOntheMind, Shewizzy2005, Cari.tenang, Maria Johansson, Alexandre.Morgaut, Gf uip, EmausBot, John of Reading, WikitanvirBot, Booler80,
Minimac’s Clone, Phiarc, SvetCo, Germanviscuso, ClueBot NG, Ki2010, Danim, Secured128, Razorbliss, Compfreak7, Pradiq009, Matspca, Snow Blizzard, Osiris, Eduardofeld, Khiladi 2010, Cyberbot II, FlyingPhysicist, André Miranda Moreira, DallasClarke, Rzicari,
Monkbot and Anonymous: 210
• NoSQL Source: Contributors: AxelBoldt, Maury Markowitz, Jose Icaza, Pnm,
Kku, Komap, Phoe6, Ronz, Ntoll, Ehn, Timwi, Furrykef, Phil Boswell, Bearcat, Peak, Dilbert, (:Julien:), Tagishsimon, Gadfium, Coldacid,
Alexf, Beland, Euphoria, Clemwang, Rfl, MMSequeira, Smyth, Leigh Honeywell, Russss, Stephen Bain, Thüringer, Walter Görlitz,
Markito~enwiki, Bhaskar, PatrickFisher, YPavan, Eno~enwiki, Crosbiesmith, Marasmusine, Woohookitty, Linas, Tshanky, Barrylb,
Dm~enwiki, Tabletop, MacTed, Nileshbansal, BD2412, Qwertyus, Koavf, Ceefour, Strait, Amire80, Seraphimblade, ErikHaugen, Professionalsql, Vegaswikian, Jubalkessler, ElKevbo, Dmccreary, AlisonW, RobertG, Sstrader, Intgr, Tedder, Benbovee, Wavelength, Hairy
Dude, Bovineone, Morphh, SamJohnston, Mbonaci, Rjlabs, Leotohill, Poohneat, GraemeL, Volt42, HereToHelp, Jonasfagundes, JLaTondre, Shepard, Matt Heard, Benhoyt, A bit iffy, SmackBot, Fulldecent, Anastrophe, Mauls, Drttm, Gorman, Somewherepurple, KiloByte,
Thumperward, Jstplace, Jerome Charles Potts, Милан Јелисавчић, Frap, DavidSol, Cybercobra, Plustgarten, Looris, ThomasMueller,
Trbdavies, NickPenguin, Eedeebee, ThurnerRupert, Petr Kopač, Zaxius, Lguzenda, Heelmijnlevenlang, Omidnoorani, Mauro Bieg, Benatkin, Mjresin, Hu12, Charbelgereige, Dancrumb, Gpierre, Arto B, Raysonho, Sanspeur, Ostrolphant, ProfessorBaltasar, Netmesh, OmerMor, Neustradamus, ColdShine, Mydoghasworms, Viper007Bond, Headbomb, CharlesHoffman, Peter Gulutzan, Davidhorman, Philu,
Bramante~enwiki, Nick Number, Sorenriise, Polymorph self, Widefox, QuiteUnusual, Replizwank, Lfstevens, Gstein, Syaskin, Dericofilho, Joolean, Orenfalkowitz, Kunaldeo, Kgfleischmann, Philg88, Mitpradeep, Adtadt, GimliDotNet, Lmxspice, Stimpy77, Mikek999,
DatabACE, JohnPritchard, Ansh.prat, McSly, Atropos235, Lamp90, Jottinger, Anoop K Nayak, Bbulkow, Tonyrogerson, Robert1947,
Rogerdpack, Billinghurst, Quiark, Kbrose, ManikSurtani, TJRC, Dawn Bard, Whimsley, DavidBourguignon, Flyer22 Reborn, Hello71,
Ctxppc, Mesut.ayata, Legacypath, AndrewBass, Edlich, Drq123, CaptTofu, Stevedekorte, Rossturk, Niceguyedc, Cnorvell, Pointillist,
Excirial, Zapher67, PixelBot, Dredwolff, Robhughadams, Arjayay, Razorflame, StanContributor, Irmatov, Shijucv, The-verver, Tgrall,
Miami33139, XLinkBot, Phoenix720, Duncan, Fiskbil, Whooym, Techsaint, Addbot, Fmorstatter, Mortense, Mabdul, MrOllie, LaaknorBot, Chrismcnab, Alexrakia, Getmoreatp, Luckas-bot, Yobot, Amirobot, Pcap, Ebalter, Ma7dy, AnomieBOT, Angry bee, Fraktalek,
White gecko, Materialscientist, Xtremejames183, Cyril Wack, Jabawack81, El33th4x0r, Gkorland, Tomdo08, Ubcule, ChristianGruen,
FontOfSomeKnowledge, Rtweed1955, Omnipaedista, Sduplooy, Shadowjams, Ciges, Cekli829, Sdrkyj, FrescoBot, Nawroth, Ashtango,
Sae1962, Thegreeneman5, David Paniz, Ertugka, Chenopodiaceous, Winterst, I dream of horses, Leegee23, Hoo man, Natishalom,
Michael Minh, Seancribbs, Jandalhandler, Craigbeveridge, Cnwilliams, Colemala, Argv0, Justinsheehy, AdityaKishore, Javalangstring,
Voodootikigod, JnRouvignac, Svesterli, Violaaa, Hoelzro, Magnuschr, Extrovrt101, Wyverald, Jeffdexter77, Uhbif19, Zond, Asafdapper, Ptab, Tobiasivarsson, Alexandre.Morgaut, Steve03Mills, Phunehehe, R39132, EmausBot, Biofinderplus, WikitanvirBot, FalseAxiom, Bdijkstra, Dewritech, GoingBatty, RA0808, Ledhed2222, MrWerewolf, EricBloch, Hloeung, ZéroBot, Weimanm, Al3xpopescu,
Theandrewdavis, Mhegi, Sagarjhobalia, Mtrencseni, Phillips-Martin, Dmitri.grigoriev, H3llBot, Jnaranjo86, DamarisC, Dstainer, Bulwersator, Eco schranzer, Thomas.uhl, Lyoshenka, Inmortalnet, Really Enthusiastic, Germanviscuso, Stephen E Browne, ClueBot NG, Rabihnassar, Ki2010, Randl, Luisramos22, Fxsjy, Korrawit, Tylerskf, Castncoot, ScottConroy, Jrudisin, Mshefer, Ashtango5, Helpful Pixie
Bot, Pereb, William greenly, Rpk512, GlobalsDB, DBigXray, Tuvrotya, BG19bot, Nawk, Gonim, Freshnfruity, Vychtrle, Gaborcselle,
Kkbhumana, Frze, AvocatoBot, Mark Arsten, Compfreak7, Anne.naimoli, Matspca, Boshomi, Fceller, Dshelby, Brocsima, BigButterfly, Winston Chuen-Shih Yang, Griswolf, Socialuser, Ugurbost, BattyBot, Khiladi 2010, Noah Slater, Farvartish, Knudmoeller, Electricmuffin11, Mbarrenecheajr, Corrector623, Sandy.toast, F331491, Luebbert42, Holland.jg, Anujsahni, Tsvljuchsh, Makecat-bot, Fitzchak,
Toopathfind, Msalvadores, Cloud-dev, Sasindar, Zhanghaohit, CJGarner, Crosstantine, Stevenguttman, Razibot, DallasClarke, Altered
Walter, Rediosoft, Tsm32, François Robere, Harpreet dandeon, LeeAMitchell, Mbroberg, Virendervermamca, Anilkumar1129, Mhgrove, FranzKraun, Jasonhpang, Nanolat, Nosql.analyst, Rzicari, Ginsuloft, Sugamsha, K0zka, Tshuva, Dodi 8238, Dabron, Mongochang,
Natan.puzis, Webtrill, CafeNoName, Yasinaktimur, Monkbot, Itamar.haber, User db, Columbus240, Textractor, Maykurutu, Kamaci,
Mongodbheavy, Dexterchief, RedOctober13, Nathan194, Jjrenzel, Azanebrain, Annmkelly1, Sunnyeyre, Danny996, Thomas4019, Dr.
Testificate, M.D., Johnbweeks, Magic-carpet-pilot, Heymattallen, Teowey, IdlePlayground, Datadatadatadata, Kevin at aerospike, Anywhichway, Nawazdhandala, The King Breaker, Mohammad.rafigh, Mgentz, Desertrat1969 and Anonymous: 551
• Key-value database Source: Contributors: Maury Markowitz,
Bearcat, Beland, ArnoldReinhold, MZMcBride, Kgfleischmann, FrescoBot, Danos p, GGink, Altered Walter, Itamar.haber, Opencooper
and Anonymous: 6
• Document-oriented database Source: Contributors:
Maury Markowitz, Edward, Ehn, RedWolf, Gwicke~enwiki, Cobaltbluetony, Chris Wood, Beland, Plasma east, Iznogoud~enwiki, Thorwald, Rfl, Imroy, Enric Naval, Shenme, Mdd, Arthena, Stuartyeates, Woohookitty, Mindmatrix, Dm~enwiki, Kingsleyj, MassGalactusUniversum, JIP, Dmccreary, Crazycomputers, Intgr, Sderose, KirtWalker, Wavelength, MySchizoBuddy, Cedar101, Thumperward,
Jerome Charles Potts, Frap, Cybercobra, Superjordo, Eedeebee, Vivek.raman, FatalError, Spdegabrielle, Cydebot, Philu, QuiteUnusual,
Dasfrosty, Plamoa, Toutoune25, Nyttend, R'n'B, Nwbeeson, Mqchen, Vishal0soni, TXiKiBoT, Kulkarninikhil, Rachkovsky, Benrcass,
Antony.stubbs, Niceguyedc, Pointillist, Boleyn, Lodrian~enwiki, Addbot, Mortense, Fgnievinski, SDSWIKI, Goldzahn, Bunnyhop11, Pcap,
Carleas, AnomieBOT, FreeRangeFrog, Neitherk, ChristianGruen, GrouchoBot, Rtweed1955, FrescoBot, Mark Renier, BrideOfKripkenstein, Bablind, RedBot, Refactored, Argv0, Mreftel, Hchrm, EmausBot, Akagel, EricBloch, Bxj, Staszek Lem, ClueBot NG, Rezabot,
Danim, JasonNichols, Helpful Pixie Bot,, Mark Arsten, Compfreak7, Luebbert42, Dodilp, CJGarner, Crosstantine, Altered
Walter, Rediosoft, Hzguo, Tsm32, François Robere, Mbroberg, Cbuccella, Tshuva, ScotXW, Dodi 8238, There is a T101 in your kitchen,
Webtrill, Adventurer61, Heymattallen, Datadatadatadata, ChrisChinchillaWard, Haptic-feedback, Jorl17, Oleron17, Tannerstirrat and
Anonymous: 106
• NewSQL Source: Contributors: Maury Markowitz, Beland, Julian
Mehnle~enwiki, Stuartyeates, MacTed, Quuxplusone, Intgr, Datamgmt, Amux, Frap, Apavlo, Kgfleischmann, Ibains, Duncan.Hull,
MPH007, Phoenix720, Dsimic, MrOllie, Yobot, AnomieBOT, Noq, W Nowicki, Diego diaz espinoza, Cnwilliams, Bulwersator, UMDDatabase, BG19bot, Akim.demaille, Dexbot, Plothridge, Sanketsarang, Brianna.galloway, Mwaci99, Andygrove73, Monkbot, Sergejjurecko, Qid4475, Hvaara, Oleron17 and Anonymous: 19
• ACID Source: Contributors: AxelBoldt, Verloren, Matusz, PierreAbbat, Fubar Obfusco, Maury Markowitz, Zippy, Mrwojo, Edward, Michael Hardy, Kku, Markonen, Karada, Haakon, Poor Yorick, IMSoP, Clausen,
GregRobson, Mcenedella, Zoicon5, Robbot, Kristof vt, RedWolf, Bernhard Bauer, Rfc1394, DHN, Jleedev, Zigger, Leonard G.,
Kainaw, Neilc, Beland, Saucepan, Vina, Daniel11, Urhixidur, Rfl, Discospinster, Rich Farmbrough, Thomas Willerich, Ponder, Elwikipedista~enwiki, Rlaager, Smalljim, Shenme, LuoShengli, Raja99, BlueNovember, Espoo, Anthony Appleyard, Suruena, Endersdouble, Ceyockey, Forderud, UFu, Mindmatrix, Swamp Ig, Barrylb, Kam Solusar, WadeSimMiser, Turnstep, Yurik, Rjwilmsi, Salix alba,
Raztus, FayssalF, FlaBot, StephanCom, Ysangkok, Kmorozov, Fragglet, Quuxplusone, Intgr, Joonasl, Chobot, YurikBot, Personman,
Pip2andahalf, Petiatil, Mskfisher, Barefootguru, Rsrikanth05, Ytcracker, Jpbowen, CPColin, Larsinio, RUL3R, Jessemerriman, MacMog,
Paul Magnussen, [email protected], Saeed Jahed, Gorgan almighty, Katieh5584, Benandorsqueaks, Victor falk, KnightRider~enwiki,
BonsaiViking, SmackBot, Amolshah, Renku, Bmearns, Mcherm, Gilliam, Thumperward, Prachee.j, DHN-bot~enwiki, Decibel, CorbinSimpson, Grover cleveland, SeanAhern, Luís Felipe Braga, Acdx, Dave.excira, Accurizer, Bezenek, IronGargoyle, Lee Carre, TwistOfCain, Paul Foxworthy, MrRedwood, Jontomkittredge, FatalError, SqlPac, Ivan Pozdeev, Safalra, Christian75, DumbBOT, Surturz, Viridae,
Thijs!bot, Epbr123, Marek69, Vertium, Uiteoi, Jaydlewis, Siggimund, Hmrox, AntiVandalBot, Gioto, Seaphoto, Lfstevens, Stangaa, DAllardyce, MetsBot, R'n'B, Tgeairn, Huzzlet the bot, J.delanoy, Trusilver, Inimino, It Is Me Here, Gurchzilla, NewEnglandYankee, DorganBot,
Tagus, Inter16, WhiteOak2006, Reelrt, Jeff G., Af648, Drake Redcrest, Sean D Martin, Martin451, BwDraco, Noformation, Duncan.Hull,
Wykypydya, Zhenqinli, Charliearcuri, Triesault, Synthebot, !dea4u, Sesshomaru, Heiser, YonaBot, Kaell, Flyer22 Reborn, JCLately, Svick,
AlanUS, Siskus, Denisarona, Loren.wilton, ClueBot, Jagun, Boing! said Zebedee, Passargea, Gakusha, Excirial, SoxBot III, Trefork, Trvth,
Maimai009, Addbot, Some jerk on the Internet, Ngpd, Shmuelsamuele, CanadianLinuxUser, Download, CarsracBot, Tide rolls, Yobot,
Synchronism, AnomieBOT, ThinkerFeeler, Jim1138, Kd24911, Nmfon, Flewis, Materialscientist, RobertEves92, Citation bot, E2eamon,
Dudegroove, Vhabacoreilc, Obersachsebot, Pontificalibus, Miym, Cole2, Tabledhote, Tct13, FrescoBot, Bluiee, Mark Renier, MGA73bot,
HJ Mitchell, Sae1962, DivineAlpha, Citation bot 1, Bunyk, Redrose64, I dream of horses, Hellknowz, SpaceFlight89, Σ, Throwaway85,
Vrenator, Premsurya, Noommos, Gf uip, DASHBot, RA0808, Tommy2010, TuHan-Bot, Wikipelli, John Cline, Fæ, Sckolar, Makecat,
Prasannawikis, Tolly4bolly, Rob7139, Puffin, Wildrain21, ClueBot NG, Andrei S, 123Hedgehog456, Widr, Ghostdood, Novusuna, Strike
Eagle, Titodutta, Calabe1992, Lowercase sigmabot, Jordonbyers, Amiramix, MPSUK, Mark Arsten, Silvrous, Yourbane, Agnt9, Klilidiplomus, Fylbecatulous, Winston Chuen-Shih Yang, Yazan kokash23, Thecodysite1, Mdann52, Polupolu890, Arr4, ZappaOMati, EuroCarGT,
Kirilldoom16, Harsh 2580, Dexbot, Golfguy399, Mynameissskr, NewAspen1, Epicgenius, Woo333, 7Rius, DevonDBA, 7Sidz, Midget
zombie, Kethrus, Winghouchan, Mayank0001, Bryanleungnokhin12345 and Anonymous: 489
• Consistency (database systems) Source: Contributors:
Patrick, Greenrd, Rfl, CanisRufus, Suruena, Ewlyahoocom, Intgr, YurikBot, Ste1n, Sasuke Sarutobi, Gaius Cornelius, SmackBot, Silly
rabbit, Capmo, Gregbard, StudierMalMarburg, Jeepday, JCLately, M4gnum0n, Addbot, Willking1979, Fraggle81, Obersachsebot, Mark
Renier, Sae1962, SpaceFlight89, Rob7139, Encyclopedant, Monkbot, Ganeshdtor, Nerdgonewild and Anonymous: 15
• Durability (database systems) Source: Contributors:
Edward, CesarB, Clausen, LordHz, Tobias Bergemann, Saucepan, Rfl, Ewlyahoocom, SmackBot, Bluebot, Wizardman, SqlPac, Astazi,
JCLately, D3fault, Addbot, Erik9bot, Mark Renier, Yourbane, JYBot and Anonymous: 8
• Serializability Source: Contributors: Ahoerstemeier, Greenrd, DavidCary,
ArnoldReinhold, Arthena, Ruud Koot, MassGalactusUniversum, BD2412, Rjwilmsi, Darthsco, Wavelength, That Guy, From That Show!,
SmackBot, Amux, Chris the speller, Mihai Capotă, Flyguy649, Cybercobra, Paul Foxworthy, Comps, MeekMark, Paddles, Kubanczyk,
Supparluca, VoABot II, Rxtreme, R'n'B, Deor, VolkovBot, Klower, JCLately, Svick, M4gnum0n, Addbot, Fyrael, Alex.mccarthy, Zorrobot,
Luckas-bot, Yobot, AnomieBOT, Materialscientist, LilHelpa, Gilo1969, Miym, Omnipaedista, FrescoBot, Mark Renier, Craig Pemberton,
Farhikht, Tbhotch, DRAGON BOOSTER, John of Reading, Dewritech, Fæ, Mentibot, ClueBot NG, Jack Greenmaven, Richard3120,
MerlIwBot, Kgrittn, Cyberpower678, Cyberbot II and Anonymous: 50
• Isolation (database systems) Source: Contributors: AxelBoldt, Ramesh, IMSoP, Hadal, Tobias Bergemann, Beland, J18ter, Asqueella, Rfl, KeyStroke, EmmetCaulfield, Velella, Mindmatrix,
10.1. TEXT
Swamp Ig, Mattmorgan, Mandarax, Bunchofgrapes, Ketiltrout, Ej, Michal.burda, Maxal, Chris Purcell, Ewlyahoocom, Alvin-cs, Ivansoto,
RussBot, Phlip, Stefan Udrea, Laurent Van Winckel, Closedmouth, SmackBot, Khfan93, Serhio, Gracenotes, Cybercobra, Djmitche, JoeBot, Igoldste, Insanephantom, Nczempin, Ervinn, Slazenger, TheJC, Thijs!bot, Maverick13, JMatthews, Rsocol, Mentin, Cameltrader, Magioladitis, Hheimbuerger, MartinBot, Scku, NunoFerreira, SoCalSuperEagle, Wikidemon, Inovakov, Enigmaman, JCLately, Paul Clapham,
Ian Clelland, ClueBot, Erichero, The Thing That Should Not Be, Mild Bill Hiccup, Niceguyedc, LonelyBeacon, Alexbot, Addbot, Materialscientist, Xqbot, Prunesqualer, SPKirsch, Sae1962, Searcherfinder, Irbisgreif, Sahedin, BYVoid, DARTH SIDIOUS 2, QLineOrientalist,
Hrishikeshbarua, Anonymouslee,, Olof nord, Tommy2010, Olnrao, ClueBot NG, Kgrittn, BG19bot, Wiki13, Brian.low22,
Snow Blizzard, Sbose7890, Tommy0605, MaryEFreeman, Kevsteppe, KasparBot and Anonymous: 176
• Database transaction Source: Contributors: Damian Yerrick,
Nixdorf, SebastianHelm, WeißNix, Ajk, Clausen, GregRobson, Owen, Craig Stuntz, RedWolf, Babbage, KellyCoinGuy, Lysy,
Mintleaf~enwiki, T0m, Jason Quinn, Timo~enwiki, Neilc, Troels Arvin, Burschik, KeyStroke, Mike Schwartz, DCEdwards1966,
Obradovic Goran, Haham hanuka, Jeltz, Derbeth, Forderud, Mindmatrix, TigerShark, AnmaFinotera, Turnstep, OMouse, FlaBot, Dauerad,
Intgr, Karel Anthonissen, Chobot, Bgwhite, YurikBot, Matiash, Hede2000, SAE1962, Larsinio, Luc4~enwiki, Mikeblas, Adi92~enwiki,
SmackBot, Georgeryp, Gilliam, Lubos, PureRED, Khukri, MegaHasher, RichMorin, [email protected], Slakr, Paul Foxworthy, Comps, SqlPac,
WeggeBot, Stevag, Thijs!bot, CharlotteWebb, JAnDbot, Geniac, Cic, Leeborkman, Hbent, Idioma-bot, TXiKiBoT, Zhenqinli, Billinghurst,
Gerd-HH, Daniel0524, Prakash Nadkarni, BotMultichill, Roesser, JCLately, Fratrep, OKBot, ClueBot, Binksternet, DnetSvg, M4gnum0n,
Triwger, HumphreyW, Addbot, Ghettoblaster, Highguard, Sandrarossi, Lightbot, Jarble, Yobot, Pcap, AnomieBOT, Rubinbot, JackieBot,
Materialscientist, Zerksis, Pepper, Mark Renier, Al3ksk, RedBot, Lingliu07, Sobia akhtar, Gf uip, K6ka, Rocketrod1960, ClueBot NG,
MerlIwBot, Helpful Pixie Bot, Mostafiz93, Kirananils, AmandeepJ, ChrisGualtieri, Davew123, Lemnaminor, Appypani, Juhuyuta, Thisismyusername96 and Anonymous: 94
• Transaction processing Source: Contributors:
Markowitz, Zippy, Pratyeka, Clausen, GregRobson, Gutza, SEWilco, Craig Stuntz, Tobias Bergemann, Khalid hassani, Uzume, Beland,
Abdull, Gordonjcp, Atlant, Wtmitchell, Stephan Leeds, Suruena, Mindmatrix, Ruud Koot, MONGO, Mandarax, BD2412, Chobot, Cliffb,
Mikeblas, Zzuuzz, Rbpasker, Tschristoppe~enwiki, Kgf0, SmackBot, Chairman S., Agateller, BBCWatcher, Avb, Radagast83, Akulkis,
Luís Felipe Braga, Joshua Scott, [email protected], JHunterJ, Beve, Baiji, Stymiee, Adolphus79, Bruvajc, Thijs!bot, Kubanczyk, JAnDbot, MER-C,
Donsez, DGG, Gwern, MartinBot, Rettetast, DeKXer, Jmcw37, STBotD, Jeff G., Lear’s Fool, Andy Dingley, JCLately, CutOffTies,
Oxymoron83, Jan1nad, M4gnum0n, Ghaskins, Oo7nets, Addbot, MrOllie, Download, LaaknorBot, Lightbot, Wireless friend, Luckas-bot,
Yobot, Pcap, Peter Flass, AnomieBOT, Jim1138, Materialscientist, Ellynwinters, Unimath, Xqbot, Mika au, Amaury, Alkamins, Mark
Renier, Charleyrich, Danielle009, Awolski, Gf uip, Cbwash, Dr Jonx, ClueBot NG, CaroleHenson, Danim, Jorgenev, Helpful Pixie Bot,
BG19bot, Bnicolae, JoshuaChen, Wiki-jonne, Hampton11235, ClaeszXIV and Anonymous: 90
• Journaling file system Source: Contributors: Marj Tiefert, Brion
VIBBER, Wesley, Uriyan, The Anome, Tarquin, Stephen Gilbert, Enchanter, Rootbeer, Ghakko, Ark~enwiki, Hephaestos, Graue, Karada,
(, Emperorbma,, Pistnor, Furrykef, Taxman, Khym Chanur, Phil Boswell, Robbot, Scott McNay, Naddy, Tim Ivorson, Cek,
David Gerard, DavidCary, AviDrissman, Mintleaf~enwiki, AlistairMcMillan, Wmahan, Karlward, Beland, Kareeser, Damieng, Mormegil,
ChrisRuvolo, KeyStroke, Luxdormiens, Indil, Edward Z. Yang, Androo, MARQUIS111, Poli, Guy Harris, Apoc2400, Seans Potato Business, Bart133, MIT Trekkie, Ruud Koot, Anthony Borla, Graham87, Rjwilmsi, Raffaele Megabyte, FlaBot, Maxal, Chobot, The Rambling
Man, YurikBot, Wikipedia., NickBush24, Moppet65535, Nailbiter, Daleh, Eskimbot, Chris the speller, Charles Moss, Anabus, SheeEttin,
Mwtoews, Ryulong, Bitwise, RekishiEJ, Rogério Brito, Unixguy, Seven of Nine, Davidhorman, Escarbot, AntiVandalBot, Widefox, Shlomi
Hillel, Gavia immer, VoABot II, Public Menace, Cpiral, WJBscribe, DorganBot, Jcea, VolkovBot, AlnoktaBOT, Dani Groner, Steve0702,
Leopoldt, MadmanBot, Rdhettinger, PipepBot, BJ712, Subversive.sound, Dsimic, Addbot, Ghettoblaster, AkhtaBot, Luckas-bot, Ptbotgourou, AnomieBOT, FrescoBot, Mfwitten, Citation bot 1, Hajecate, Merlion444, Japs 88, 15turnsm, Aharris16, ClueBot NG, BG19bot,
Tsjerven, Forsakensam and Anonymous: 96
• Atomicity (database systems) Source: Contributors:
Michael Hardy, DopefishJustin, CesarB, Rohan Jayasekera, Biggins, Gdimitr, Hadal, Jleedev, Enochlau, Ancheta Wis, ArneBab, Rfl, Smyth,
Hooperbloob, EmmetCaulfield, Danhash, RJFJR, Apokrif, PeterJohnson, Marudubshinki, Nihiltres, Chris Purcell, Ewlyahoocom, Julescubtree, Snailwalker, Chobot, Korg, RussBot, Sasuke Sarutobi, Bota47, Vicarious, SmackBot, Jmendez, Betacommand, DanPope, TimBentley, LinguistAtLarge, Hkmaly, Kvng, Dreftymac, JForget, Neelix, Bodragon, JPG-GR, TXiKiBoT, Synthebot, Freshbaked, JCLately,
Qwertykris, Addbot, Jncraton, Fraggle81, Materialscientist, Erik9bot, Mark Renier, Sae1962, Widr, Anutural, Juan Carlos Farah, TvojaStara, Cooldudevipin and Anonymous: 37
• Lock (database) Source: Contributors: Greenrd, Ta bu shi da yu, Duplode, Caidence, Maxal, SmackBot, InverseHypercube, Mauro Bieg, VinnieCool, Bruvajc, Thijs!bot, Wikid77, Belenus, Cic, Jack007,
Jojalozzo, Kidoshisama, Addbot, LilHelpa, Vishnu2011, Gulsig4, Potionism, ClueBot NG, Danim, Sharpshooter4008, EdwardH, ChrisGualtieri, FCutic and Anonymous: 20
• Record locking Source: Contributors: Michael Hardy, Finn-Zoltan, D6,
Atlant, Pol098, SmackBot, Quaddriver, Umaguna, JCLately, Jojalozzo, Niceguyedc, Erik9bot, This lousy T-shirt, Waynelwarren, BattyBot
and Anonymous: 15
• Two-phase locking Source: Contributors: Michael Hardy, Poor
Yorick, Clausen, Dtaylor1984, Neilc, Andreas Kaufmann, Rich Farmbrough, Nchaimov, Woohookitty, Ruud Koot, Wavelength, Aaron
Schulz, SmackBot, OrangeDog, Jeskeca, Cybercobra, Comps, SeanMon, Epbr123, Touko vk, Seaphoto, Beta16, Syst3m, Lerdthenerd,
Paul20070, Gerakibot, Yintan, Svick, ImageRemovalBot, Stuart.clayton.22, Addbot, Cxz111, Thomas Bjørkan, Yobot, AnomieBOT, Materialscientist, AbigailAbernathy, Craig Pemberton, John of Reading, ClueBot NG, Chrisjameskirkham, Cntras, Helpful Pixie Bot, Dexbot,
Jodosma, JohnTB and Anonymous: 41
• Multiversion concurrency control Source: Contributors: Poor Yorick, Palfrey, Ggaughan, Dcoetzee, Jamesday, RickBeton, Craig Stuntz, Rsfinn, DavidCary, Neilc, Chowbok, Rawlife, Troels
Arvin, Rich Farmbrough, Smyth, Martpol, R. S. Shaw, Franl, Terrycojones, RJFJR, Drbreznjev, GregorB, Ssteedman, Turnstep, Rjwilmsi,
Ysangkok, Chris Purcell, Intgr, YurikBot, Piet Delport, Gaius Cornelius, AmunRa, Blowdart, Naasking, JLaTondre, That Guy, From That
Show!, SmackBot, Basil.bourque, Chris the speller, ThurnerRupert, Cbbrowne, Hu12, Tawkerbot2, ChrisCork, Comps, Raysonho, Gritzko,
Elendal, Hga, Jordan Brown, ProfessorBaltasar, Cydebot, Cwhii, Marcuscalabresus, Nowhere man, Andrewjamesmorgan, Visik, Dougher,
Tedickey, Seashorewiki, Ahodgkinson, Kiore, Yannick56, Kedawa, DanielWeinreb, Fecund, Bill.zopf, Dllahr, Arleach, Highlandsun, Dfetter, Doug4j, Danilo.Piazzalunga, MrChupon, Whimsley, Jerryobject, JCLately, Kobotbel, Breinbaas, Siskus, Tuntable, Jonathanstray,
Nthiery, CYCC, M4gnum0n, Kalotus, KyleJ1, MelonBot, XLinkBot, MystBot, Addbot, Tsunanet, Lightbot, Yobot, Pcap, Pvjohnson,
AnomieBOT, Yoonforh, Drachmae, ThomasTomMueller, LilHelpa, FrescoBot, LucienBOT, DrilBot, RedBot, Seancribbs, Full-date unlinking bot, Unordained, Obankston, John of Reading, Dewritech, EricBloch, Tuhl, H3llBot, Stradafi, Dexp, CasualVisitor, BG19bot, Wasbeer, Julien2512, Snnn~enwiki, Compfreak7, Kevin 71984, Waynelwarren, Kwetal1, Giloki, JYBot, Bdempsey64, JingguoYao, Mbautin,
Will Faught, Kevinroy09, Johnkarva, Plothridge, Textractor, Kanelai, Kclee968, Oleron17 and Anonymous: 108
• Snapshot isolation Source: Contributors: Craig Stuntz, Neilc, Isidore,
Chowbok, Woohookitty, Ruud Koot, Ej, Ysangkok, Chris Purcell, Blowdart, Elkman, Johndburger, SmackBot, Comps, Cydebot, Andrewjamesmorgan, David Eppstein, VanishedUserABC, JCLately, Tuntable, Idleloop~enwiki, Yobot, Pcap, AnomieBOT, Citation bot,
Searcherfinder, Citation bot 1, AnnHarrison, Helpful Pixie Bot, Kgrittn and Anonymous: 13
• Two-phase commit protocol Source: Contributors: Pnm,
Ciphergoth, Lkesteloot, Gtrmp, Neilc, Gdr, Rworsnop, Rdsmith4, Rich Farmbrough, CanisRufus, R. S. Shaw, Liao, Bestchai, Mbloore,
Suruena, ReubenGarrett, Ruud Koot, Choas~enwiki, MassGalactusUniversum, YurikBot, Bayle Shanks, Daleh, Segv11, Emilong, SmackBot, Dubwai, Jdeisenh, Coredesat, Zero sharp, Comps, Pmerson, Touko vk, LenzGr, Somebody2014, MartinBot, Stephanwehner, Richard
KAL, Ja 62, VolkovBot, Cyberjoac, JCLately, Svick, Yagibear, WikHead, Addbot, DOI bot, Twimoki, Yobot, AnomieBOT, Materialscientist, Uglybugger, Wktsugue, FrescoBot, Killian441, PleaseStand, EmausBot, Flegmon, ClueBot NG, Braincricket, Electriccatfish2,
BG19bot, JingguoYao, SLipRaTi, Deepu2k, Hotcheese92 and Anonymous: 76
• Three-phase commit protocol Source: Contributors:
Chris Q, Pnm, Trevor Johns, Levin, Rworsnop, N.o.bouvin, Guy Harris, Ceyockey, Ampledata, Choas~enwiki, Rjwilmsi, GreyCat, Bayle
Shanks, Aaron Schulz, Segv11, Emilong, Jsnx, Zero sharp, Megatronium, Junche, MarkKampe, TXiKiBoT, JCLately, Addbot, DOI bot,
Citation bot, FrescoBot, IdishK, Joerg Bader, BG19bot, BattyBot, Remcgrath, Monkbot and Anonymous: 26
• Scalability Source: Contributors: Kpjas, The Anome, Awaterl, Matusz,
Michael Hardy, Kku, TakuyaMurata, Bearcat, Sander123, Jondel, Dbroadwell, SpellBott, Lysy, Javidjamae, Leonard G., Stevietheman,
Gdr, Beland, Urhixidur, Hugh Mason, Ferrans, FT2, Mazi, Dtremenak, Liao, Calton, Pion, Suruena, Kusma, Mattbrundage, Tyz, Undefined~enwiki, BD2412, Rjwilmsi, Quiddity, Williamborg, Fred Bradstadt, Aapo Laitinen, FlaBot, Intgr, Dalef, Agil~enwiki, YurikBot,
Whoisjohngalt, NTBot~enwiki, Michael Slone, Bovineone, Moe Epsilon, Leotohill, .marc., Xpclient, LeonardoRob0t, Stumps, SmackBot,
Irnavash, KelleyCook, Ohnoitsjamie, Thumperward, Jammus, Javalenok, Ascentury, Frap, JonHarder, Cyhatch, BWDuncan, Andrei Stroe,
Harryboyles, Writtenonsand, [email protected], Swartik, Hu12, UncleDouggie, Tawkerbot2, FatalError, CBM, Thijs!bot, Uiteoi, Marokwitz, Kdakin,
JAnDbot, NapoliRoma, Shar1R, SunSw0rd, Raffen, Joshua Davis, FienX, RockMFR, Auroramatt, 1000Faces, NewEnglandYankee, Doria, Jottinger, Izno, VolkovBot, TXiKiBoT, CHaoTiCa, Falcon8765, Suction Man, Bpringlemeir, Paladin1979, DigitalDave42, JCLately,
Luciole2013, Gp5588, Dangelow, Nvrijn, Elnon, Tearaway, Mild Bill Hiccup, Saravu2k, M4gnum0n, Friendlydata, Shiro jdn, MPH007,
XLinkBot, Philippe Giabbanelli, Avoided, Klungel, MystBot, Dsimic, Addbot, Jncraton, Tonkie67, Fluffernutter, MrOllie, Latiligence,
Kiril Simeonovski, Teles, Luckas-bot, Yobot, Davew haverford, Terrifictriffid, AnomieBOT, Materialscientist, Obersachsebot, Xqbot,
Miym, GrouchoBot, Sae1962, MastiBot, Jandalhandler, Akolyth, Jesse V., Gf uip, EmausBot, John of Reading, Anirudh Emani, Josve05a,
Cosmoskramer, AManWithNoPlan, Music Sorter, Tsipi, ChuispastonBot, ClueBot NG, Widr, Daniel Minor, Meniv, Helpful Pixie Bot,
MarkusWinand, Electriccatfish2, BG19bot, ElphiBot, Wikicadger, Anbu121, Srenniw, BattyBot, CGBoas, K0zka, Mwaci11, Paul2520,
Shahbazali101, Igorghisi, Vieque, Verbal.noun, Mantraman701, KasparBot, The Quixotic Potato, KealanJH and Anonymous: 121
• Shard (database architecture) Source: Contributors: Rfl,
Kenfar, Fche, Gareth McCaughan, Winterstein, Bgwhite, Hairy Dude, Grafen, Neil Hooey, Wainstead, Deepdraft, SmackBot, Russorat, Cybercobra, Jdlambert, Steipe, Bezenek, Sanspeur, Underpants, Jadahl, Dougher, Magioladitis, Eleschinski2000, StefanPapp, Otisg, McSly,
NewEnglandYankee, Cswpride, Phasma Felis, Andy Dingley, Angusmca, SmallRepair, ClueBot, Fipar, Delicious carbuncle, Dthomsen8,
MystBot, Yobot, AnomieBOT, Noq, Isheden, FontOfSomeKnowledge, Rfportilla, Mopashinov, Jacosi, Tilkax, FrescoBot, X7q, Haeinous,
Sae1962, Gautamsomani, Citation bot 1, Winterst, Jandalhandler, Crysb, Visvadinu, Cfupdate, EmausBot, ClueBot NG, Liran.zelkha,
BG19bot, Eric-vrcl, MusikAnimal, MeganShield, Tmalone22, BattyBot, Jaybear, Drewandersonnz, Cc4fire, David9911, Dudewhereismybike, Alexandre.marini, Sodomojo, Prabhusingh25, Viam Ferream, Josephidziorek, Codingalz and Anonymous: 76
• Optimistic concurrency control Source: Contributors:
Karada, Poor Yorick, Nikai, Timwi, Zoicon5, DavidCary, Neilc, Beland, D6, Rich Farmbrough, Smyth, Dmeranda, Mnot, R. S. Shaw,
Bhaskar, Suruena, HenryLi, Simetrical, GregorB, Intgr, Ahunt, SAE1962, SmackBot, Slamb, SmartGuy Old, Drewnoakes, NYKevin, Allan McInnes, Russorat, Cybercobra, Iridescent, Zero sharp, Nczempin, Mydoghasworms, Lfstevens, Magioladitis, JeromeJerome, Algotr,
Homer Landskirty, Randomalious, SieBot, JCLately, DraX3D, OKBot, Svick, Jfromcanada, ClueBot, Sim IJskes, Methossant, M4gnum0n,
Addbot, Ace of Spades, BenzolBot, Citation bot 1, RedBot, RjwilmsiBot, Helpful Pixie Bot, Noazark, Therealmatbrown, Monkbot,
Bhutani.ashish14 and Anonymous: 36
• Partition (database) Source: Contributors: Ehn, Peak, Foonly, S.K.,
Stevelihn, Alai, Geoffmcgrath, Mindmatrix, Drrngrvy, Roboto de Ajvol, Yahya Abdal-Aziz, Brian1975, Mikeblas, Georgewilliamherbert,
Doubleplusjeff, Ccubedd, Salobaas, Andrew.rose, JAnDbot, Mdfst13, Jamelan, Andy Dingley, Jonstephens, Angusmca, Ceva, SmallRepair,
Pinkadelica, Saravu2k, Chrisarnesen, Addbot, Vishnava, Highflyerjl, MrOllie, SamatBot, Yobot, AnomieBOT, Beaddy1238, Materialscientist, Semmerich, Isheden, LucienBOT, Mark Renier, Wordstext, Troy.frericks, Habitmelon, Fholahan, Lurkfest, Jan.hasller, Vieque and
Anonymous: 32
• Distributed transaction Source: Contributors: Gtrmp,
Saucepan, MartinBiely, Exceeder~enwiki, Ruud Koot, Jameshfisher, X42bn6, ShinyKnows, Gaius Cornelius, SmackBot, Brick Thrower,
Jeskeca, Ligulembot, Comps, Pmerson, Darklilac, GL1zdA, Wikiisawesome, JCLately, Alexbot, MelonBot, MystBot, Addbot, Pcap,
AnomieBOT, John of Reading, K6ka, Helpful Pixie Bot, Mediran and Anonymous: 15
• Redis Source: Contributors: Thebramp, Ehn, Toreau, Beland, ShakataGaNai, Jaybuffington, Stesch, Rfl, Rich Farmbrough, Plest, Philipp Weis, CONFIQ, Barrylb, Bhound89, Justin Ormont, Sdornan, Raztus, Husky, Intgr,
Bgwhite, Iamfscked, Rsrikanth05, SamJohnston, Joel7687, Unforgiven24, Poohneat, Mike Dillon, Modify, Nic Doye, SmackBot, Larry
Doolittle, Boul22435, Wolph~enwiki, Vid, Frap, ThurnerRupert, Heelmijnlevenlang, Vanished user ih3rjk324jdei2, Aaaidan, Sanspeur,
Cydebot, Ivant, Jamitzky, Avi4now, Wdspann, HazeNZ, Jm3, Omarkonsul, Sorenriise, Gioto, Kavehmb, Adys, Scorwin, Vishal0soni,
Satani, TXiKiBoT, BlackVegetable, Jonknox, Benclewett, Swillison, ImageRemovalBot, Arkanosis, M4gnum0n, Miami33139, Dthomsen8, Rreagan007, Addbot, Mortense, Jncraton, Gnukix, Tnm8, Balabiot, Jarble, Pcap, Kmerenkov, Dmarquard, BastianVenthur, Mavz0r,
Xqbot, Soveran, Cole2, Jder~enwiki, FrescoBot, Ksato9700, Tóraí, Hoo man, Tim1357, Jfmantis, EmausBot, T. Canens, Spf2, Angrytoast, Djembayz, ZéroBot, Overred~enwiki, Corb555, Mark Martinec, Sitic, Adamretter, Kasirbot, Karmiq, Codingoutloud, BG19bot,
10.1. TEXT
Toffanin, Anne.naimoli, Fceller, Axule, Samusman waz, Rashidusman82, User 9d3ffb53ccf204e0, Ju2ender, Dexbot, Rwky, Altered Walter, Luhnatic, Dme26, Aeon10, Nikunjsingh01, Napy65, AlexanderRedd, Wksunder, Ebr.wolff, Mindskt, Mmoreram, Itamar.haber, AbinashMishra1234, Ushnishtha, S4saurabh12, Nikkiobjectrocket, Andlpa63, Terrencepryan and Anonymous: 126
• MongoDB Source: Contributors: AxelBoldt, William Avery, Phoe6, Grendelkhan, Topbanana, Tomchiukc, Wjhonson, Sriehl, Jason Quinn, Masterhomer, Coldacid, Pgan002, Beland, Alobodig, Josephgrossberg, ShortBus, Thorwald, Perey, Rfl, Rich Farmbrough, Jakuza~enwiki, Nabla, Theinfo, Lauciusa, Stesmo, Kfogel, Koper, Mdd,
PCJockey, Zachlipton, Tobych, Kocio, Nforbes, Falcorian, LOL, Adallas, Dm~enwiki, GregorB, Johnny99, VsevolodSipakov, HV, AlisonW, RobertG, Garyvdm, Kolbasz, FrankTobia, Wavelength, Hairy Dude, Youngtwig, SamJohnston, Thunderforge, Ben b, Arthur Rubin, Fram, ViperSnake151, Narkstraws, AndrewWTaylor, DrJolo, SmackBot, Mauls, KennethJ, Ohnoitsjamie, Skizzik, Jdorner, Ctrlfreak13, Frap, Meandtheshell, Ochbad, DMacks, KDIsom, Stennie, Heelmijnlevenlang, Chickencha, Hu12, MikeWazowski, Mikeyv,
Alexey Feldgendler, Mineralè, Cydebot, Calorus, Widefox, Kohenkatz, Avleenvig, Magioladitis, JamesBWatson, AlbinoChocobo, David
Eppstein, SBunce, Cander0000, Jackson Peebles, Lmxspice, CommonsDelinker, Qweruiop321, Grshiplett, BrianOfRugby, Serge925,
TXiKiBoT, BookLubber, KickaXe, Wingedsubmariner, Crcsmnky, Tuxcantfly, Valio bg, Koryu Obihiro, Iapain wiki, Quiark, Kbrose,
Adm.Wiggin, Yashwantchavan, Jojalozzo, Elibarzilay, Hello71, Svick, Solprovider, Shruti14, Gian-Pa, WDavis1911, Niceguyedc, ArlenCuss, Alexbot, Plaes, Jinlye, Supa Z, SteveMao, Megaltoid, Shijucv, The-verver, Crypticbird, Piratemurray, XLinkBot, Gerhardvalentin, MystBot, Deineka, Addbot, Moosehadley, CanadianLinuxUser, MrOllie, Jasper Deng, Peridon, Twimoki, TundraGreen, Ben
Ben, Zyx, Luckas-bot, Yobot, JackPotte, Ptbotgourou, Amirobot, Wonderfl, AnomieBOT, Beaddy1238, Jim1138, Materialscientist,
CoMePrAdZ, Cababunga, John Bessa, Gkorland, Flying sheep, Mackrauss, Miym, Jonas AGX, Omnipaedista, Yadavjpr, SassoBot, Ciges,
Mike2782, Mu Mind, Haeinous, Jocelynp85, Marsiancba, Jandalhandler, Mdirolf, Chris Caven, LogAntiLog, OnceAlpha, Mreftel, Reaper
Eternal, Difu Wu, Asafdapper, RjwilmsiBot, Mbferg, Pengwynn, WikitanvirBot, Najeeb1010, GoingBatty, Mixmax99, Al3xpopescu,
Bernard.szlachta, Shuitu, Blr21000, Lateg, H3llBot, Vittyvk, Dstainer, Zephyrus Tavvier, Petrb, ClueBot NG, Mechanical digger, Thepaul0,
Joefromrandb, Millermk, Ninja987, Renatovitolo, Mschneido, Masssly, Antiqueight, Nadavidson, Helpful Pixie Bot, BG19bot, Arkroll,
Compfreak7, Dustinrodrigues, Cnevis, Ycallaf, Tobias.trelle, Dr. Coal, Jayadevp13, Solved009, Kizar, FeralOink, Chip123456, Ethefor, Justincheng12345-bot, Bpatrick001, Shaksharf, The Illusive Man, ChrisGualtieri, JYBot, Thenaberry1, Lan3y, Dexbot, Awesoham,
Frosty, MartinMichlmayr, Samarthgahire, Jamesx12345, Mikewbaca, Alfre2v, CJGarner, DavidPKendal, CCC2012, Mahbubur-r-aaman,
Amritakripa, Fc07, François Robere, Jan.hasller, The Herald, Francium1988, Ewolff42, ArmitageAmy, Tshuva, Xujizhe, Mongodb,
Grouik92, Luxure, Kathir04on, ScotXW, Jameswahlin, Lvmetrics, Bhpdownloads, Mehdi2305, Dodi 8238, Dabron, Mongochang, Samohtm, Kaimast, Tazi Mehdi, Ayush3292, Mcocohen, Airuleguy, RationalBlasphemist, Swoopover, Vwsuran, Attish~enwiki, Alexeybut,
Andlpa63, Andreea Panainte, Dwtkr, Iamvikas1982, Xsenechal, Anandchandak15, Mandge.rohit, Rnmandge and Anonymous: 342
• PostgreSQL Source: Contributors: 0, Carey Evans, Wesley, The Anome,
Christopher Mahan, Aldie, Fubar Obfusco, Nate Silva, M~enwiki, Roadrunner, Maury Markowitz, TomCerul, Heron, Cwitty, Frecklefoot, Edward, Fuzzie, Gregben~enwiki, Nixdorf, Liftarn, Wwwwolf, (, Greenman, CesarB, Ahoerstemeier, KAMiKAZOW, Stevenj,
Nanshu, Angela, Glenn, Sugarfish, Nikai, Ehn, Jay, Tejano, Doradus, Pedant17, Tpbradbury, Polyglot, Cleduc, Jnc, Wernher, Bevo, Joy,
Fvw, Jamesday, Robbot, Chealer, Lowellian, Ianb, Stewartadcock, LX, Lasix, Weialawaga~enwiki, Oberiko, Nickdc, Levin, Fleminra,
Ceejayoz, Prell, Lvr, Neilc, Chowbok, Utcursch, Pgan002, Alexf, Cbraga, ConradPino, Billposer, Gene s, Burgundavia, Karl-Henner,
Cynical, Troels Arvin, GreenReaper, Deleteme42, RandalSchwartz, Imroy, Rich Farmbrough, Sesse, Oska, Ardonik, Lulu of the LotusEaters, Slipstream, Gronky, Bender235, PaulMEdwards, Chibimagic, Kwamikagami, Kanzure, TommyG, Mereman, Giraffedata, Timsheridan, Minghong, Jonsafari, QuantumEleven, Stephen Bain, HasharBot~enwiki, Poli, Gary, Fchoong, Jeltz, Andrewpmk, Ringerc~enwiki,
Pjacklam, Pauli133, SteinbDJ, Gmaxwell, Mindmatrix, MianZKhurrum, Dandv, Deeahbz, Jacobolus, Distalzou, Ruud Koot, Cosmicsoftceo, Bowman, GregorB, AnmaFinotera, Wisq, Turnstep, Marudubshinki, Graham87, EdDavies, Gilesmorant, KublaChaos, Silvestre
Zabala, Zero0w, Mikecron, FlaBot, SchuminWeb, Gurch, Ghen, Intgr, Chobot, Reetep, Peterl, YurikBot, Manop, Gaius Cornelius, Bovineone, Varnav, Geertivp, Mipadi, Tkbwik, Randolf Richardson, Larsinio, E rulez, Snarius, BraneJ, Gsherry, Analoguedragon, Johndburger, Closedmouth, Johnsu01, MaNeMeBasat, BanzaiSi, JLaTondre, Benhoyt, Mlibby, A bit iffy, SmackBot, Nicolas Barbier, Direvus,
Slamb, Faisal.akeel, Reedy, Georgeryp, Dkusnetzky, Anastrophe, Richmeister, Amux, Chris the speller, Bluebot, Wolph~enwiki, Advorak,
DStoykov, Crashmatrix, Ben.c.roberts, Thumperward, Droll, Jerome Charles Potts, DHN-bot~enwiki, Decibel, Frap, Lantrix, Matchups,
MattOates, Stevemidgley, Cybercobra, Emmanuel JARRI, Mwtoews, Where, Towsonu2003~enwiki, SashatoBot, Vincenzo.romano, Brian
Gunderson, Avé, Misery~enwiki, Abolen, AdultSwim, MTSbot~enwiki, Peyre, DagErlingSmørgrav, Angryxpeh, Dark ixion, Hu12, FatalError, Raysonho, WeggeBot, Musashi1600,, Revolus, Cydebot, Krauss, Ttiotsw, Rchoate, Synergy, Ebrahim,
Thijs!bot, MinorEdits, Andyjsmith, Dalahäst, Jcarle, Xzilla, AntiVandalBot, Bearheart, Marokwitz, LenzGr, Room813, Deflective, Cjkporter, Martinkunev, Dskoll, GregorySmith, Apostrophyx, NoDepositNoReturn, JamesBWatson, Bernd vdB~enwiki, Gabriel Kielland, EagleFan, Cander0000, Nevit, Gwern, Scottmacpherson, Ronbtni, Lmxspice, Zeus, R'n'B, Whale plane, Sven Klemm~enwiki, Keesiewonder,
Usp, Eleven81, VolkovBot, Allencheung, Blindmatrix, Bramschoenmakers, Luke Lonergan, Kmacd, Gwinkless, HLHJ, RonaldDuncan,
Dfetter, Eljope, Jibun, PieterDeBruijn, Agentq314, Anas2048, Kindofoctarine, Majeru, ChrisMiddleton, Bfcase, X-Fi6, Jojalozzo, Gorgot, Ctkeene, Ctxppc, Jonlandrum, Trevorbrooks, Martarius, ClueBot, Jhellerstein, Kl4m, Tintinobelisk, The Thing That Should Not Be,
Unbuttered Parsnip, Kl4m-AWB, Nikolas Stephan, Niceguyedc, M4gnum0n, WikiNickEN, Freerangelibrarian, Pmronchi, Eeekster, TobiasPersson, Chrisarnesen, SF007, DumZiBoT, TimTay, XLinkBot, Jabberwoch, Andy318, Addbot, Mortense, Nate Wessel, Ginzel~enwiki,
Fale, Luckas-bot, Yobot, Specious, AnomieBOT, Coolboy1234, Piano non troppo, Kukushk, ArthurBot, The Banner, Wikante, Gilo1969,
Zenaan, FChurca, Mark Renier, W Nowicki, Alex.ryazantsev, Kwiki, Louperibot, William.temperley, Simple Bob, B3t, Tim baroon,
Skyerise, Jandalhandler, Full-date unlinking bot, Bmomjian, Jons2006, Filiprem, Ravenmewtwo, RjwilmsiBot, Streapadair, EmausBot,
Grldedwrdbutler, Marzalpac, Klenot, Dewritech, Peaceray, Solarra, Your Lord and Master, AvicBot, H3llBot, Demonkoryu, Gz33, கி.
கார்த்திகேயன், Sbmeirow, Palosirkka, Eggyknap, Martinmarques, Tijfo098, ChuispastonBot, Rugh, Brew8028, DisneyG, ClueBot NG,
Birkedit, Mangal ratna, Redneb33, Boria, Web20boom, Kasirbot, Patrias, Denys.Kravchenko, Kwetal1, Kweetal nl, Winston Chuen-Shih
Yang, Dexbot, Codename Lisa, Laurenz albe, Palmbeachguy, Digoal, Praemonitus, RaphaelQS, Comp.arch, ScotXW, Lavagnino, Craigkerstiens, Erilong, Unician, Docesam, Simonriggs, X3mofile, Lucazeo, Helldalgo, E1328167, Johnlgrant, Vihorny, GCarterEDB, Transfat0g
and Anonymous: 408
• Apache Cassandra Source: Contributors: Enchanter, Frecklefoot,
Edward, Ronz, Stefan-S, Ehn, Hashar, Cleduc, Bearcat, Alan Liefting, Msiebuhr, Neilc, Pgan002, Beland, Euphoria, Rich Farmbrough,
Bender235, Anthony Appleyard, PatrickFisher, YPavan, Runtime, Swaroopch, Mindmatrix, Timendum, Deansfa, Qwertyus, Jmhodges,
Jweiss11, Vegaswikian, Intgr, Tas50, FrankTobia, SamJohnston, Formina Sage, Mipadi, Grafen, Wainstead, JLaTondre, Tommymorgan, Chris Chittleborough, AtomCrusher, SmackBot, Timoey, Aardvark92, Jdorner, Thumperward, OrangeDog, Frap, Cybercobra, Mwtoews, Acdx, Daniel.Cardenas, ArglebargleIV, Cydebot, Mblumber, Cinderblock63, Fyedernoggersnodden, Anupam, Nemnkim, Gstein,
GreyTeardrop, Cander0000, Krotty, McSly, Woodjr, Mercurywoodrose, Bcantoni, Andy Dingley, Peter.vanroose, Drmies, Alexbot, Arjayay, The-verver, XLinkBot, Kolyma, Maximgr, Deineka, Addbot, Mortense, S4saurabh, Jncraton, MrOllie, Mdnahas, Jbryanscott,
Luckas-bot, Yobot, Midinastasurazz, JackPotte, Vanger13, AnomieBOT, Jim1138, Materialscientist, Pmiossec, Citation bot, ArthurBot, Xqbot, Santiagobasulto, Yadavjpr, Cgraysontx, Ciges, FrescoBot, FalconL, IO Device, Ksato9700, Sae1962, Tdmackey, Freenerd, Jadave234, Biktora, Rollins83, DASHBot, Dewritech, GoingBatty, Peaceray, Driftx, Arthurjulian, Werieth, ZéroBot, Al3xpopescu,
Kylemurph, Grossenhayn, Dstainer, Billmantisco, ClueBot NG, Nmilford, Ben morphett, Elisiariocouto, Kinglarvae, Helpful Pixie Bot,
YogiWanKenobi, Slebresne, BG19bot, Mark Arsten, IluvatarBot, Andrewllavore, Mmozum, Diglio.simoni, BattyBot, RichardMills65,
Perlscriptgurubot, ChrisGualtieri, TheJJJunk, Viocomnetworks, Clydewylam, Hoestmelankoli, Samarthgahire, Jamesx12345, CJGarner,
RobinUS2, Jonathanbellis, Wikiuser298, Jimtarber, Stuartmccaul, Stather, Mfiguiere, Dough34, Sebastibe, Khyryll, Dabron, Samohtm,
Spiesche, Textractor, Wehanw, FLGMwt, Elishaoren, Andlpa63, Jericevans, Kevin at aerospike, BD2412bot, Wesko, Marissabell, Jorgebg,
Barmic and Anonymous: 172
• Berkeley DB Source: Contributors: Damian Yerrick, AxelBoldt, Bryan
Derksen, Taw, Andre Engels, Nate Silva, DavidSJ, Danja, Phoe6, KAMiKAZOW, Docu, Mxn, GregRobson, Jay, Johnleach, Smithd,
Paul W, Peak, Merovingian, Tobias Bergemann, Centrx, BenFrantzDale, AlistairMcMillan, Mendel, Rworsnop, Am088, Oneiros, Alobodig, Bumm13, Arunkv, EagleOne, RossPatterson, Rich Farmbrough, Mecanismo, Abelson, Gronky, Jarsyl, Elwikipedista~enwiki, Mike
Schwartz, TommyG, JesseHogan, Justinc, Beinsane, Lystrata, TheParanoidOne, Interiot, GregorB, Tom W.M., Tokek, Turnstep, Kesla,
Graham87, Art Cancro, Qwertyus, Nuptsey, Ligulem, Gene Wood, Zero0w, FlaBot, Vsion, Kmorozov, Intgr, Jac099, YurikBot, Rylz,
Gaius Cornelius, SamJohnston, DragonHawk, DavidConrad, Welsh, Gregburd, Echiner~enwiki, JLaTondre, Berlotti, ViperSnake151,
SmackBot, Reedy, Vald, KiloByte, Thumperward, Jerome Charles Potts, Letdorf, Eliyahu S, Where, TPO-bot, AThing, Guyjohnston,
Jeberle, Rafert, Fatespeaks, Aarktica, Riffic, Eliashc, Tawkerbot2, 00112233, FatalError, Raysonho, Cydebot, Wernight, Danarmstrong,
Gioto, Gilson.soares, Marokwitz, Mk*, Skarkkai, Jelloman, MER-C, Wootery, Ff1959, Cander0000, Gwern, Ariel., CommonsDelinker,
Lordgilman, Rivenmyst137, Rajankila, Limn, Highlandsun, UnitedStatesian, Pcolby, Jerryobject, Oerdnj, Dead paulie, Toothrot, Explicit, Mild Bill Hiccup, Kilessan, Ecobun, Feline Hymnic, GreenGourd, Rhododendrites, SoxBot, TobiasPersson, Manorhill, Paulsheer,
Brentsmith101, Drhowarddrfine, Twimoki, Legobot, Yobot, Legobot II, AnomieBOT, Arcoro01, Rubinbot, Götz, Noelkoutlis, Edrandall,
Drilnoth, Scientes, UrusHyby, FrescoBot, HJ Mitchell, Nathan 314159, GoodenM, I dream of horses, Skyerise, SchreyP, Tokmeserdar,
JnRouvignac, Figarobdbxml, McShark, Bolerio, DigitalKiwi, Ebrambot, Staticd, Mikhail Ryazanov, Dexp, Be..anyone, BG19bot, Compfreak7, Gm246, Chrisxue815, ChrisGualtieri, MartinMichlmayr, Hoestmelankoli, LennyWikidata, Apgiannakidis, ScotXW, Shinydiscoball
and Anonymous: 123
• Memcached Source: Contributors: AxelBoldt, Magnus Manske, Bryan Derksen, Tim Starling, Lemming, Angela, Den fjättrade ankan~enwiki, Ijon, Alaric, Ehn, Markhurd, Grendelkhan, Toreau, Ayucat, Rrjanbiah,
TittoAssini, Mattflaschen, David Gerard, Jpo, Christopherlin, Oneishy, Neilc, Pgan002, Fpahl, Euphoria, Beginning, Gcanyon, Thorwald,
Rfl, Rich Farmbrough, Gronky, Bender235, Enyo, CanisRufus, John Vandenberg, Giraffedata, Gary, Halsteadk, Diego Moya, GreggHilferding, Stephan Leeds, Danhash, Foolswisdom, Lime~enwiki, Simetrical, Barrylb, Jacobolus, Bratsche, Apokrif, MrSomeone, Rjwilmsi,
Pleiotrop3, Pmc, Mahlon, Jnutting512, Anrie Nord, Pinecar, Phorque, YoavShapira, Rsrikanth05, SamJohnston, Corevette, Gslin, SmackBot, Mark Tranchant, Faisal.akeel, Oben, DStoykov, Morte, Oli Filth, ClaudiaM, Frap, Hermzz, T.J. Crowder, BrentRockwood, A5b,
Evlekis, Thejerm, Sembiance, Tawkerbot2, Chnv, HDCase, Raysonho, Devis, Jasoncalacanis, Colorprobe, Grubbiv, Johnnicely, Thijs!bot,
Dasani, Plausible deniability, Cherianthomas, Spotnyk, Hcobb, Mordere, Marokwitz, Pensador82, Falcor84, Mange01, Charlesyoung, AntiSpamBot, Tarinth, HansRoht, LastChanceToBe, TXiKiBoT, AgamemnonZ, Mfriedenhagen, Zkarab, BrianAker, Cnilep, Hyuu 91, Aednichols, NightRave, Monkeychubs, SHINERJ6, Kl4m-AWB, Gtstricky, Pauljchang, Herbert1000, SF007, Qgil-WMF, Addbot, ToikeOike,
Raykrueger, Peridon, OlEnglish, Pjherrero, Legobot, Luckas-bot, Yobot, Timeroot, Kowser, Bub’s, AnomieBOT, Materialscientist,
Pmiossec, Neurocod, DataWraith, The Evil IP address, PlaysWithLife, FrescoBot, Natishalom, Enes1177, Sevi81, Jfmantis, Ingenthr,
Majidkhan59, EmausBot, SBachenberg, Ajraddatz, Wikipediancag, ZéroBot, ILYA INDIGO, Javiermarinros, Zephyrus Tavvier, Smcquay, ClueBot NG, Webkul, Danim, Kronigon1, Helpful Pixie Bot, Mhmhk, Compfreak7, CitationCleanerBot, OmarMOthman, Papowell, JYBot, Dexbot, Altered Walter, RaphaelQS, Bitobor, ArmitageAmy, Mindskt, ScotXW, Dudeprgm, Kevin at aerospike, Gekart and
Anonymous: 171
• BigTable Source: Contributors: Jpatokal, Hashar, Greenrd, Bearcat, Bkonrad,
Jason Quinn, Adpowers, Macrakis, Khalid hassani, Pgan002, Jeremykemp, Imroy, Discospinster, Brianhe, Pmsyyz, Bender235, Nileshbansal, Marudubshinki, Zoz, Rjwilmsi, Vegaswikian, Intgr, Wavelength, Cliffb, Markpeak, Corevette, Moe Epsilon, DeadEyeArrow, Elkman, Georgewilliamherbert, Matt Heard, SmackBot, Gnangarra, Drttm, Hmains, Amux, Fintler, JennyRad, Jeskeca, Audriusa, Cybercobra,
Dwchin, Nakon, EIFY, A5b, Michael miceli, Larrymcp, Skim1420, John Reed Riley, Raysonho, Phoib, Franzks, Harrigan, Oo7565, Cydebot, Nearfar, Heroeswithmetaphors, Gstein, Benstown, Exerda, Gwern, Erik s paulson, Sounil, Rdquay, Vincent Lextrait, TXiKiBoT,
Andy Dingley, Treekids, Dstrube, Captin411, ArlenCuss, Alexbot, Whereiswally, Replysixty, XLinkBot, Addbot, Tanhabot, Luckas-bot,
Yobot, Legobot II, Wonderfl, Masharabinovich, AnomieBOT, ArthurBot, Quebec99, DataWraith, Ogendarf, Thehelpfulbot, Mateuszb,
FrescoBot, Mark Renier, Sae1962, Wordstext, RedBot, MastiBot, Barauswald, Trappist the monk, Jmspaggi, DanielWaterworth, Pirroh,
Ebrambot, FinalRapture, ClueBot NG, Dfarrell07, Barry K. Nathan, Danim, FreePeter3000, BG19bot, Mgr493, AvocatoBot, Goolasso,
BattyBot, Paul2520, Monkbot, Seano314, Aashrayarora, Jefesaurus and Anonymous: 99
10.2 Images
• File:ASF-logo.svg Source: License: Apache License 2.0 Contributors: Original artist: Apache Software Foundation (ASF)
• File:Ambox_important.svg Source: License: Public domain Contributors: Own work, based off of Image:Ambox scales.svg Original artist: Dsmurat (talk · contribs)
• File:BabbageKeyValueCard.tiff Source: License:
Public domain Contributors:, https:
// Original artist: Charles Babbage
• File:CO-ScheduleClasses.jpg Source: License: Attribution Contributors: Yoav Raz (1990): The Principle of Commitment Ordering, or Guaranteeing Serializability in a Heterogeneous Environment of Multiple Autonomous Resource Managers Using Atomic Commitment. DEC-TR 841, Digital Equipment Corporation, November
1990; Yoav Raz (1995): The Principle of Commitment Ordering, in Original artist: Yoav Raz
10.2. IMAGES
• File:CodasylB.png Source: License: CC-BY-SA-3.0 Contributors:
“CIM: Principles of Computer Integrated Manufacturing”, Jean-Baptiste Waldner, John Wiley & Sons, 1992 Original artist: Jean-Baptiste
• File:Commons-logo.svg Source: License: ? Contributors: ? Original
artist: ?
• File:Computer-aj_aj_ashton_01.svg Source:
Yellow_theme.svg License: CC0 Contributors: Original artist: AJ from
• File:Crystal_Clear_app_database.png Source:
png License: LGPL Contributors: All Crystal Clear icons were posted by the author as LGPL on kde-look; Original artist: Everaldo Coelho
and YellowIcon;
• File:Database_models.jpg Source: License: CC BY-SA
3.0 Contributors: Own work Original artist: Marcel Douwe Dekker
• File:Disambig_gray.svg Source: License: Cc-by-sa-3.0 Contributors:
? Original artist: ?
• File:Edit-clear.svg Source: License: Public domain Contributors: The
Tango! Desktop Project. Original artist:
The people from the Tango! project. And according to the meta-data in the file, specifically: “Andreas Nilsson, and Jakub Steiner (although
• File:Folder_Hexagonal_Icon.svg Source: License: Cc-bysa-3.0 Contributors: ? Original artist: ?
• File:Free_Software_Portal_Logo.svg Source:
License: LGPL Contributors:
• Nuvola_apps_emacs.png Original artist: Nuvola_apps_emacs.png: David Vignoni
• File:Gnome-searchtool.svg Source: License: LGPL Contributors: Original artist: David
• File:Helenos_for_Apache_Cassandra.PNG Source:
Cassandra.PNG License: CC BY-SA 4.0 Contributors: Own work Original artist: JackPotte
• File:HelloWorld.svg Source: License: Public domain Contributors: Own work Original artist: Wooptoo
• File:Internet_map_1024.jpg Source: License: CC BY
2.5 Contributors: Originally from the English Wikipedia; description page is/was here. Original artist: The Opte Project
• File:LampFlowchart.svg Source: License: CC-BY-SA-3.0
Contributors: vector version of Image:LampFlowchart.png Original artist: svg by Booyabazooka
• File:Memcached.svg Source: License: Fair use Contributors:
The logo may be obtained from Memcached.
Original artist: ?
• File:Merge-arrow.svg Source: License: Public domain Contributors: ? Original artist: ?
• File:Merge-arrows.svg Source: License: Public domain Contributors: ? Original artist: ?
• File:Mergefrom.svg Source: License: Public domain Contributors: ? Original artist: ?
• File:MongoDB-Logo.svg Source: License: Fair use Contributors:
The logo is from the website. Original artist: ?
• File:Node.js_logo.svg Source: License: Public domain Contributors: Original artist: node.js authors
• File:Object-Oriented_Model.svg Source: License:
Public domain Contributors: Data Integration Glossary. Original artist:
• U.S. Department of Transportation
• vectorization: Own work
• File:Office-book.svg Source: License: Public domain Contributors: This and myself. Original artist: Chris Down/Tango project
• File:People_icon.svg Source: License: CC0 Contributors: OpenClipart Original artist: OpenClipart
• File:Portal-puzzle.svg Source: License: Public domain Contributors: ?
Original artist: ?
• File:Postgresql_elephant.svg Source: License: BSD
Contributors: Original artist: Jeff MacDonald
• File:Question_book-new.svg Source: License: Cc-by-sa-3.0
Created from scratch in Adobe Illustrator. Based on Image:Question book.png created by User:Equazcion Original artist:
• File:Redis_Logo.svg Source: License: Fair use Contributors: http:
// Original artist: Carlos Prioglio created the logo for the copyright owner Salvatore Sanfilippo, lead developer
of Redis.
• File:Relational_key_SVG.svg Source: License: CC
BY-SA 3.0 Contributors: Own work Original artist: IkamusumeFan
• File:Robomongo_0.8.5_-_insertion.png Source:
insertion.png License: CC BY-SA 4.0 Contributors: Own work Original artist: JackPotte
• File:Star-schema-example.png Source: License: CC-BYSA-3.0 Contributors:
I created this work entirely by myself.
Original artist:
SqlPac (talk)
• File:Symbol_book_class2.svg Source: License: CC
BY-SA 2.5 Contributors: Mad by Lokal_Profil by combining: Original artist: Lokal_Profil
• File:Text_document_with_red_question_mark.svg Source:
with_red_question_mark.svg License: Public domain Contributors: Created by bdesham with Inkscape; based upon Text-x-generic.svg
from the Tango project. Original artist: Benjamin D. Esham (bdesham)
• File:Three-phase_commit_diagram.png Source:
License: CC-BY-3.0 Contributors: ? Original artist: ?
• File:Traditional_View_of_Data_SVG.svg Source:
Data_SVG.svg License: CC BY-SA 3.0 Contributors: Own work Original artist: IkamusumeFan
• File:Unbalanced_scales.svg Source: License: Public domain Contributors: ? Original artist: ?
• File:Wikibooks-logo.svg Source: License: CC BY-SA 3.0
Contributors: Own work Original artist: User:Bastique, User:Ramac et al.
• File:Wikinews-logo.svg Source: License: CC BY-SA 3.0
Contributors: This is a cropped version of Image:Wikinews-logo-en.png. Original artist: Vectorized by Simon 01:05, 2 August 2006 (UTC)
Updated by Time3000 17 April 2007 to use official Wikinews colours and appear correctly on dark backgrounds. Originally uploaded by
• File:Wikiquote-logo.svg Source: License: Public domain
Contributors: ? Original artist: ?
• File:Wikisource-logo.svg Source: License: CC BY-SA 3.0
Contributors: Rei-artur Original artist: Nicholas Moreau
• File:Wikiversity-logo-Snorky.svg Source: License:
CC BY-SA 3.0 Contributors: Own work Original artist: Snorky
• File:Wikiversity-logo.svg Source: License: CC BY-SA 3.0
Contributors: Snorky (optimized and cleaned up by verdy_p) Original artist: Snorky (optimized and cleaned up by verdy_p)
• File:Wiktionary-logo-en.svg Source: License: Public
domain Contributors: Vector version of Image:Wiktionary-logo-en.png. Original artist: Vectorized by Fvasconcellos (talk · contribs),
based on original logo tossed together by Brion Vibber
10.3 Content license
• Creative Commons Attribution-Share Alike 3.0
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF