White Paper
Extreme Performance Using Oracle TimesTen
In-Memory Database
ORACLE WHITE PAPER
|
JUNE 2014
Table of Contents
Introduction
3
Milliseconds Matter
3
The Growth of Real-Time Applications
3
Real-Time Industries
3
Real-Time Enterprises
4
Real-Time Data Management Software
4
Application-Tier Deployment
4
Oracle TimesTen In-Memory Database
4
In-Memory Database Technology
5
The Physical Structures of Oracle TimesTen
6
Application-Tier Shared Libraries
6
Memory-Resident Data Structures
7
System Processes
7
Administrative Programs
7
Checkpoint and Log Files
8
Data Replication Technology
8
Caching Technology
8
IMDB Technology in Depth
9
Query Optimization
9
Indexing
10
Query Processing
10
Buffer Pool Management
10
1 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
The Benefits of the Difference
11
A New Look at Price Performance
11
Exceptional Performance
11
Scalability
11
Response Time
12
Real-Time Functionality
13
Data Management
13
Durability and Concurrency
14
Disk Operations
14
Locking and Isolation
15
Query Processing
15
Data Replication
16
Balancing Performance and Consistency
16
Replication Topologies
17
Caching
18
Oracle TimesTen Application-Tier Database Cache
18
Event Processing
19
Conclusion
2 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
20
Introduction
The Oracle TimesTen In-Memory Database is a memory-optimized relational database that delivers
very low response time and very high throughput for performance-critical systems. It is targeted to run
in the application tier, close to applications, and optionally in process with applications. It can be used
as the database of record or as a cache to the Oracle Database.
This paper describes the Oracle TimesTen In-Memory Database, and gives a high-level description of
how it may be used as a cache to the Oracle Database. For a more complete description of how to
use TimesTen as a cache to the Oracle Database, see1.
Milliseconds Matter
Oracle TimesTen provides application-tier data management for performance-critical systems, and is optimized for
blazing-fast response and real-time caching of Oracle data. Companies can extend their software infrastructures
with Oracle TimesTen to create systems that are instantly responsive, highly scalable, and continuously available.
These systems are used to increase customer loyalty, attract new customers, streamline operations, and avoid the
costly alternative of proprietary software development
Oracle TimesTen has a proven track record, with production deployments since 1998 in real-time enterprises and
time-critical environments such as network telecommunication services, operational support systems, contact
centers, airline and reservation systems, command and control systems, and securities trading. Thousands of
companies worldwide use Oracle TimesTen in production applications—including Alcatel-Lucent, Aspect, Avaya,
Bank of America Merrill Lynch, Bridgewater Systems, BroadSoft, Cisco, Deutsche Börse, Ericsson, JPMorgan,
KDDI, NEC, NYFIX, Smart Communications, United States Postal Office and Verizon Wireless.
The Growth of Real-Time Applications
Network equipment manufacturers, telecom operators, securities exchanges and brokerages, airlines, shipping and
logistics companies, and defense and intelligence agencies are examples of enterprises in which real-time
applications are a necessity. The use of real-time processing to capture, analyze, and respond intelligently to key
events is increasingly becoming the benchmark for corporate excellence.
Real-Time Industries
For many companies, real-time applications aren’t elective—they are a necessity. Network equipment
manufacturers, telecom operators, securities exchanges and brokerages, airlines, shipping and logistics companies,
and defense and intelligence agencies are prime examples of enterprises that require real-time applications. In the
past, building these applications also required the development of real-time infrastructure software. Fast but
inflexible, these systems did the job as long as the applications remained static in their requirements. But dynamic
industries quickly outpace static applications, and the cost to develop, test, and maintain specialized infrastructure
software is rarely justified when commercial alternatives exist.
1 Using Oracle TimesTen Application-Tier Database Cache to Accelerate the Oracle Database. An Oracle White Paper, June 2014
3 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Real-Time Enterprises
With the growing speed of messages moving through business networks, the use of real-time processing to capture,
analyze, and respond intelligently to key events is becoming the benchmark for corporate excellence. This isn’t
important only for the execution and management of critical business processes. Customers expect highly tailored
interactions and the utmost responsiveness from any company with which they do significant business.
Business activity monitoring, complex event processing, RFID/sensor-based applications, Web portals, and Web
services are contributing to the movement of applications to the edge of the enterprise. Configured as dynamic
collections of interrelated components, these applications are part of an overall approach known as a serviceoriented architecture (SOA). However, most data sources still reside in the back office, dominated by large amounts
of rarely touched legacy data surrounding smaller amounts of currently active information. A natural extension of
SOA concepts includes lightweight, real-time data management in the application tier, connected to corporate data
sources to provide real-time performance for currently active data.
Real-Time Data Management Software
The enterprise architectures that derive the greatest benefit from real-time processing provide event, data, and
transaction management in the application tier, empowering front-line systems with rapid response and deeper
insight. It is not sufficient to merely collect and cache data next to applications, as is often the case with firstgeneration in-house efforts. Nor is it practical to locate the corporate database on the same platform as one of the
applications.
What’s needed is a generation of lightweight infrastructure software that presents familiar, powerful interfaces and
query languages that are widely in use—that can easily interface with existing back-office databases, messaging
systems, and application servers—and that exploits the full performance potential of today’s networked, memoryrich computing platforms. This is the generation of infrastructure software for real-time data management provided
by the Oracle TimesTen In-Memory Database.
Application-Tier Deployment
Much of today’s new application development is focused on improving interactions with customers and streamlining
internal operations to eliminate delays and excess costs. These are real-time applications that reside near the
network edge or, in some cases, within the network as hosted services.
This is the new enterprise application tier, with different platform, performance, and availability requirements than
legacy back-office applications. Business events are embodied in network messages to which applications
subscribe, triggering real-time processing and additional message publishing.
To meet the response time and scalability goals for these applications, it’s often necessary to deploy the
infrastructure software and applications on the same platform, including some or all the data that drives the
application.
Oracle TimesTen is designed to integrate seamlessly into these environments, with an architecture that is optimized
for application-tier deployment and configuration options that enable highly tailored solutions.
Oracle TimesTen In-Memory Database
Oracle TimesTen In-Memory Database (TimesTen) is a memory-optimized relational database that delivers to
applications the instant responsiveness and very high throughput required by real-time enterprises and industries.
4 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
TimesTen is typically deployed in the application tier. TimesTen databases fit entirely in physical memory. They are
persistent and recoverable, and access to them is provided via standard SQL interfaces.
TimesTen databases may be replicated between servers for high availability and load sharing. Data replication
configurations can be set to active/standby or active/active, using asynchronous or synchronous transmission, with
conflict detection and resolution, and automatic failover, recovery, and resynchronization after a failed server is
restored.
The Oracle TimesTen Application-Tier Database Cache (TimesTen Cache) is the pre-integrated SQL database
cache option to the Oracle Database. It caches real-time, updatable subsets of an Oracle database in TimesTen in
the application tier. It automatically synchronizes data between the caches and the Oracle database. It offloads
computing cycles from back-end Oracle Database servers and enables remarkably responsive and scalable realtime applications. The TimesTen Cache automates pass-through of SQL requests for non-cached data, and
automatically resynchronizes data after failures.
In-Memory Database Technology
In-memory database technology implements a relational database in which all data at runtime resides in the RAM.
The data structures and access algorithms exploit this property for breakthrough performance.
In-memory database (IMDB) technology is the foundation technology for TimesTen. IMDB technology implements a
relational database in which all data at runtime resides in RAM, and the data structures and access algorithms
exploit this property for breakthrough performance. Compared to a fully cached RDBMS, IMDB technology requires
far less processing power, because the overhead to manage memory buffers and account for multiple data locations
(disk and memory) is eliminated. With IMDB technology, disks are used for persistence and recovery rather than as
the primary database storage location.
The memory-optimized performance of TimesTen is complemented by transactional properties, persistence
mechanisms, and recovery from system failures. A variety of choices is available for locking, multiuser isolation, and
logging, accommodating a range of application scenarios from transient look-up caches to core transactional trading
and billing systems.
Durability is achieved by logging the changes from committed transactions to disk and periodically updating a disk
image of the database, termed a “checkpoint.” The timing of the disk write for the log is configurable by the
application; it can be either synchronous with the end of the transaction or deferred until after. Many situations favor
higher throughput over synchronous logging, particularly when the monetary value of a transaction is low or the
transaction data is short-lived, such as when tracking the location of mobile phones that communicate their cell
location every few seconds.
The interfaces supported by TimesTen are standards-compliant. Applications issue SQL and PL/SQL commands
using ODBC, JDBC, ADO.NET, the Oracle Call Interface (OCI), or TTClasses, as well as Pro*C. TimesTen C++
Interface Classes (TTClasses) is a C++ class library that provides wrappers around the most common ODBC
functionality. It is easier to use than ODBC and promotes best practices while maintaining fast performance. The
statements for defining databases, replication configurations, and cache groups also adhere to SQL syntax
conventions. Simple network management protocol is used to issue standardized system management alerts.
5 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
An open transaction log API (XLA) with a standard Java message service (JMS) interface is provided for reading the
transaction log. This is useful for creating applications that react to database updates. In this regard, XLA is a
lightweight trigger. It’s also a way to build custom data replication from Oracle TimesTen products to other database
systems.
The Physical Structures of Oracle TimesTen
This section describes the system components of TimesTen. TimesTen consists of
» Shared libraries
» Memory-resident data structures
» System processes
» Administrative programs
» Checkpoint files and log files on disk
Application-Tier Shared Libraries
The components that implement the functionality of TimesTen are embodied in a set of shared libraries that
developers link to their applications and execute as a part of the application’s process. This shared library approach
is in contrast to a more conventional RDBMS, which is implemented as a collection of executable programs that
applications connect to, typically over a client/server network.
Normally, embedding data manager libraries into the application could make the database vulnerable to corruption if
the application process were terminated abnormally. TimesTen solved this challenge. Using a patented algorithm
known as MicroLogging, TimesTen libraries self-protect against application process failures. In-memory databases
remain consistent, and other applications continue without impact.
Figure 1. Components of the Oracle TimesTen In-Memory Database
6 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
IMDB technology is implemented as in-memory databases accessed by applications, utility programs, and system
processes via shared library routines. Disk files for logs and checkpoints (backup copies) are maintained for
recovery purposes.
Applications may use a client/server connection to access an Oracle TimesTen database, though in most cases the
best performance will be realized with a directly linked application.
Memory-Resident Data Structures
In-memory databases are maintained in an operating system’s shared memory segments, and contain all user data,
indexes, system catalogs, log buffers, lock tables, and temp space. Multiple applications can share a TimesTen
database, and a single application can access multiple TimesTen databases on the same system.
Memory-resident databases are maintained in an operating system’s shared memory segments and contain user
data, indexes, system catalogs, log buffers, lock tables, and temp space. Multiple applications can share a
database, and a single application can access multiple databases.
Application Program
Application
code
TimesTen
TimesTen
shared
libraries
shared libraries
In-Memory
Database
In-Memory
Database
In-Memory
Database
Data Tables,
Indexes,
System Tables
Locks, Cursors,
Compiled Commands,
Temp Indexes
Log
Buffer
Temporary
Memory
Region
Permanent Memory
Region
Figure 2. Oracle TimesTen in-memory database
System Processes
Background processes provide services for startup, shutdown, and application failure detection at the system level,
and provide loading, checkpointing, and deadlock handling at the database level. There is one instance-wide
TimesTen daemon (a system may have multiple instances, i.e., multiple installations of the TimesTen software), and
a separate subdaemon for each database.
Administrative Programs
Utility programs are explicitly invoked by users, scripts, and applications to perform services such as interactive
SQL, bulk copy, backup/restore, database migration, and system monitoring.
7 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Checkpoint and Log Files
Changes to the database and transaction logs are written to disk periodically. Should a database need to be
recovered, TimesTen merges the database checkpoint on disk with the completed transactions that are still in the
log files. Normal disk file systems are used for checkpoints and log files.
Data Replication Technology
When high availability or workload distribution is desired, data replication can be configured to send updates
between two or more servers. A master server is configured to send updates, and a subscriber server is configured
to receive them, and a server can be both a master and a subscriber for bidirectional replication. Time-based conflict
detection and resolution is used to establish precedence in the rare event of the same data being updated in
multiple locations at the same time.
When replication is configured, a replication agent process is started for each database. If multiple databases on the
same server are configured for replication, there is a separate replication agent for each database. Each replication
agent can send updates to one or more subscribers, and receive updates from one or more masters. Each of these
connections is implemented as a separate thread of execution inside the replication agent process. Replication
agents communicate through TCP/IP stream sockets.
For maximum performance, the replication agent detects updates to a database by monitoring the existing
transaction log, and sends updates to the subscribers in batches if possible. Only committed transactions are
replicated. On the subscriber node, the replication agent updates the database though an efficient low-level
interface, avoiding the overhead of the SQL layer.
Caching Technology
When TimesTen Application-Tier Database Cache (TimesTen Cache) is used to cache portions of an Oracle
Database, one or more cache groups are created to hold the cached data, and a system agent (the cache agent)
performs all asynchronous data transfers between the cache and Oracle Database.
A cache group is a collection of one or more tables arranged in a logical hierarchy via primary/foreign key
relationships. Each table in a cache group is related to an Oracle Database table. A cache group table can contain
all or a subset of the rows and columns in the related Oracle Database table. Cache groups can be created and
modified via SQL and PL/SQL statements. Cache groups support the following features:

Applications can both read from and write to tables in the cache groups.

Cache groups can be refreshed from an Oracle database automatically or manually.

Updates to cache groups can be propagated to an Oracle database automatically or manually.

Changes to either Oracle Database tables or cache groups can be tracked automatically.
When rows in a cache group are updated by applications, the corresponding rows in Oracle Database tables are
updated synchronously as part of the same transaction, or asynchronously immediately afterward depending on the
type of cache group that was created. Changes that originate in Oracle Database are refreshed into the cache via
the cache agent.
Cache Grids are available for horizontal scalability in performance and capacity where a Cache Grid consists of a
collection of TimesTen Caches that collectively manage an application’s cached data. Cached data is distributed
between grid members, and the Cache Grid provides applications with location transparency. Cache Grids enable
8 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
incremental scalability through the online addition (and removal) of grid members. They maintain consistency of
cached data between the Cache Grid members and the Oracle Database.
2
For more information on the TimesTen Application-Tier Database Cache, see .
IMDB Technology in Depth
By managing all data in memory and optimizing for that environment, in-memory database technology can operate
much more efficiently, and thus offer dramatic improvements in performance.
In-memory database technology delivers impressive performance by changing the assumptions about where data
resides at runtime. By managing all data in memory and optimizing for that environment, in-memory database
technology can operate much more efficiently, and thus offer dramatic improvements in responsiveness and
throughput.
But what’s at the heart of this performance? Couldn’t an application get the same results just by placing all its
RDBMS data in main memory?
Much of the work that is done by an RDBMS is done under the assumption that data is primarily on disk. Oracle
TimesTen, on the other hand, assumes the data resides in main memory and can therefore take more direct routes
to it, reducing code-path length and simplifying both algorithm and structure.
In comparing these architectures, there are a number of key differences that clearly illustrate how a many-fold
performance improvement is attainable. To illustrate, just a few of these differences can be found in query
optimization algorithms, indexing, query processing, and buffer pool management.
Query Optimization
Because disk input/output is far more expensive than memory access, disk-based RDBMSs have to assume that
data resides on disk.
Query optimization algorithms are different for disk-based systems than for memory-based systems. RDBMS
optimization decisions are based on the assumption that data resides primarily on disk. In a dynamic runtime
environment, data might be on disk or cached in main-memory at any given moment. Because disk input/output
(I/O) is far more expensive than memory access, disk-based RDBMSs have to reduce the probability of access to
disk as much as possible. They are optimized to reduce the point of performance bottleneck, so a disk-based
optimizer will not always produce the optimal plan for data that resides—primarily or fully—in main memory.
IMDB technology, on the other hand, knows that the data resides in main memory and optimizes its queries under
simpler assumptions. It does not need to make a worst-case scenario based on disk residency, and therefore, its
cost estimates can be simpler and more consistently accurate.
2 Oracle TimesTen Application-Tier Database Cache to Accelerate the Oracle Database. An Oracle White Paper, June 2014
9 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Indexing
Indexes are as important to the performance of memory-based systems as they are for disk-based systems.
Indexes are used for a variety of purposes during query processing, such as quickly qualifying rows in a SQL
SELECT, UPDATE or DELETE statement, finding the minimum or maximum value of a given column or set of
columns, or speeding up join processing. Indexes can also provide an ordered stream of records during query
processing. Such ordered streams can eliminate the need for additional sort operations, or allow for fast duplicate
elimination.
For all of these similarities with indexes in disk-based RDBMs, indexes can have significant differences with IMDB
technology. Where each index entry in a disk-based RDBMS index usually contains an index key and a record
identifier, which enables the RDBMS to locate the record on disk, with IMDB technology, keys do not need to be
stored in the indexes, and record identifiers can be implemented as record pointers. A record pointer in an index
entry points to the corresponding record that contains the key. By avoiding the duplication of key values in the index
structure, IMDB technology greatly reduces the size of each index. The absence of key values can also lead to a
simpler index implementation because it avoids the need to manage variable-size keys within the index nodes.
Query Processing
IMDB technology takes advantage of memory residency of all data. For example, the fastest way to execute a
query with an ORDER BY clause in a disk-based RDBMS is if the base table happens to be stored in sorted order
using the same column as in the ORDER BY clause. That is of course very unlikely. The next best case is if the
order can be satisfied using an index scan. However, retrieving the data via an index scan results in random access
to the records, and potentially results in very high I/O cost. It may even require that the same data block be read
from disk multiple times. Since the base table can only be stored in sorted order via one set of columns, the order in
which the records are stored can be only used to satisfy ORDER BY queries where the ORDER BY columns are a
prefix of the sorting columns. This is not an issue with IMDB technology because index entries point directly to the
memory address where the data resides, so random scan via an index is as cheap as a sequential scan.
Similarly, when implementing a sort function in a disk-based system, one has to decide whether the sort structure
will contain entire or partial records, key/record identifier pairs, or only record identifiers. This is a
space/performance tradeoff. If the attributes of interest are not stored in the sort structure, random access with
potential I/O overhead will be needed to retrieve the data. By contrast, memory-based systems do not have this
issue because data can be accessed directly via tuple pointers so the sort structure needs to contain only the tuple
pointers.
Buffer Pool Management
Although necessary in disk-based data management solutions, buffer pools become unnecessary in memory-based
data management because the data already resides within memory.
In conventional RDBMS architectures, buffer pools must be maintained for data that has been cached in main
memory. When the query processor requires a page of data, it must first search the buffer pool for that data, and
even if the data is there, in many cases it must be copied out of the pool for processing. This buffer pool
maintenance and management, coupled with additional data copies, add significantly to the original burden of
making the data available to an application.
10 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Although necessary for disk-based databases, buffer pools are unnecessary for in-memory databases. The data
already resides within main memory, so TimesTen has no buffer pool. Therefore, code path length and engine
footprint are reduced, copying is avoided, algorithms are simplified, and data is delivered to the application more
quickly.
The Benefits of the Difference
When the assumption of data residing on disk (or rather, the ambiguity of data location) is removed, complexity is
dramatically reduced. The number of machine instructions drops considerably, buffer pool management disappears,
extra data copies aren’t needed, index pages shrink, and their structure is simplified. When data residing in memory
is the bedrock assumption, everything gets simpler, more compact, and faster.
A New Look at Price Performance
For many applications, leveraging in-memory database technology offers a new way to gain a competitive edge,
improve user satisfaction, and boost return on investment.
Performance improvements can be measured in multiples rather than fractions. An application that spends half its
processing time managing data and half its time within its application space or business logic can nearly double its
capacity by using an in-memory data management product. This dramatically increases the options in terms of
market strategies, architectural economies, and system choice.
Exceptional Performance
Scalability
TimesTen takes advantage of multiple CPUs on symmetric multiprocessor computers. Figure 3 shows the
transaction throughput of TimesTen on an Intel E5-2680 @2.7GHz 2 sockets 8cores/socket system running Oracle
Linux. Each transaction executes a single SQL select (read) or update operation, as indicated. The results are
shown for 1, 4, 8 and 16 concurrent processes.
Figure 3. The throughput of TimesTen
11 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
The simplest and therefore highest volume workload is the “100% reads” test, which measures how many individual
records can be retrieved using a key value lookup. With 16 concurrent processes, 6,050,188 reads were completed
every second, on average. With only one process, the read rate was 421,795 per second.
Update operations require more processing than read operations, partly because changes must be logged for
recovery purposes. In this result, more than 952,778 records were updated per second with 16 concurrent
processes, and more than 130,375 with a single process.
Oracle TimesTen products are capable of exceptional throughput, ranging from tens of thousands of update
transactions to hundreds of thousands of read operations per second, when the transaction contains a single SQL
operation.
Although these pure workloads aren’t representative of any particular business application, these results validate the
extreme efficiency of TimesTen, and any realistic mixture of SQL operations would likely result in a maximum
throughput in the tens-of-thousands-per-second range. If higher volumes are required, TimesTen scale well beyond
a 16-way system. Note that performance results vary on different processor platforms.
Response Time
Throughput is a byproduct of individual transaction response times. As throughput increases, average response time
decreases. The exceptional throughput of Oracle TimesTen products yield microsecond-level response times on
systems across a range of data management workloads. The average response times for the throughput workloads
previously described are shown in Figure 4.
Oracle TimesTen products’ response time is measured in microseconds for single-record operations. Data-intensive
applications will benefit the most, since overall response time for the application includes more than just the data
management portion.
Figure 4. Latency in Oracle TimesTen products
It’s important to keep in mind that a transaction’s response time is the aggregate of the response times of the
constituent processing steps, such as business logic, networking, I/O, and data management. The performance of
TimesTen’s portion is not the only determinant of overall response time. Sometimes one of the steps, usually I/O or
networking, will dominate the others.
12 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
For instance, if client/server network is used instead of a direct connection between the application and the data
manager, an additional millisecond or more of response time can be added for each round trip over the network.
Similarly, if synchronous I/O is used for logging changes, the relative slowness of a disk operation is included in the
response time. In both cases, if the data management component is completed within microseconds, the overall
application response time is still going to be a factor of the network and I/O latencies.
For applications that have a dominant data management component, the speedup of TimesTen technology could
achieve a major decrease in response time. The highest-performing application is one that is directly connected,
able to use asynchronous logging, and is data management–intensive. Search-intensive online reservation systems,
presence, and location-based communication services and contact center routing engines are good examples.
As always, performance results vary greatly depending on many factors, such as the size of a transaction, the
amount of non-data access logic, and platform performance.
Real-Time Functionality
This section presents a functional view of the Oracle TimesTen In-Memory Database, describing the interfaces and
features used by developers when writing real-time applications.
Data Management
The data manager feature in TimesTen implements a true relational database model. It is not a hybrid model or a
different model with a relational view grafted on top. Developers who are familiar with popular RDBMS products
should be immediately productive.
Relational Database Model
In the relational model, user data is stored as rows (records) within tables. A row is a list of columns (data fields),
each of a specified data type. Indexes are explicitly or implicitly created to speed key-based or value-range searches
of the tables. Oracle TimesTen products support both hash indexes and range indexes. Defined constraints, such as
uniqueness or referential integrity (parent/child relationships between tables), are enforced automatically.
Within a table, a primary key—comprised of one or more columns—may be designated. A primary key is the unique
identifier for a row, and Oracle TimesTen automatically prevents the insertion of additional rows with the same
primary key value. One or more foreign keys may also be designated per table, each associated with the primary
key of another table. Foreign keys enforce a parent/child relationship between two tables. For example, a row with a
primary key will not be deleted if an associated foreign key in another table has the same value.
A view is a virtual table whose content consists of the result of the query that makes up the view definition. Views
can be included in queries just as ordinary tables would be.
A materialized view is a read-only table derived from regular (base) tables. It can include parts or all of the data from
one or more tables, and can include calculated values (such as totals and averages). Since materialized views are
automatically updated by the system whenever necessitated by a change to a base table, they provide applications
with faster access to frequently derived data. Without materialized views, the overhead of joins and other
calculations would be incurred whenever an application needed to read the data. Materialized views are especially
useful in support of event processing (discussed in more detail later in this article).
13 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
If the TimesTen Cache option is used, one or more cache groups can be defined in an in-memory database. A
cache group is a hierarchical set of tables (enforced through foreign keys) that map to a corresponding set of Oracle
Database tables, and contain some or all of the same data. Cache groups are both read and write, and updates to
the data can propagate automatically between TimesTen and the Oracle Database.
Data management operations are expressed via the industry-standard Structured Query Language (SQL), and via
PL/SQL and Pro*C. Tables, indexes, views, materialized views, and cache groups are created and maintained
within the databases. Applications connect to a TimesTen database or to a TimesTen Cache database and perform
operations against the data within. Multiple databases may exist on the same computer system and be accessed by
the same application. However, no single operation (for example, a join across two tables) may span more than one
database. Applications can participate in a distributed transaction through TimesTen’s support of the standard XA
(or JTA for Java) interface for two-phase commit.
Durability and Concurrency
TimesTen conforms to the atomicity, consistency, isolation, and durability properties of data management systems.
TimesTen transactions conform to the atomicity, consistency, isolation, and durability properties. These properties
ensure that, in a multiuser system, each transaction operates as if it were the only transaction being executed at the
time, and that the system can guarantee that the effects of a committed transaction are not lost. These are the most
rigid properties required of data managers, and TimesTen ensures full conformance.
A common misperception is that in-memory data managers cannot prevent data loss from system failures. In fact,
the same techniques that make transactions and data durable in a conventional database are used in TimesTen. As
in all transaction-oriented systems, durability is achieved through a combination of change logging and periodic
refreshes of a version of the database that resides on a disk.
TimesTen offers applications control as a tradeoff for the degree of durability and the total throughput. More of one
means less of the other, and TimesTen provides choices along this spectrum.
Disk Operations
Oracle TimesTen databases may be permanent or temporary, and for exclusive or shared access.
If an application requires that no changes be lost, log records are flushed to disk as part of committing the
transaction. If maximum performance is more important than possibly losing some transactions, log records can be
written to disk less often, asynchronously from each transaction commit. In either case, TimesTen will attempt to
“group commit” multiple transactions together to minimize disk writes.
Periodically, the database is checkpointed to disk automatically based on user-specified frequency and log volume.
TimesTen supports “fuzzy” checkpoint operations that run in the background with minimal impact on running
applications. Recovery from a system failure is a matter of merging log records with the latest checkpoint file.
TimesTen databases may be permanent or temporary. Permanent databases (checkpoint files) remain on disk
when not in use. Temporary databases reside solely in memory and are removed when not in use. Permanent
databases are the most commonly deployed configurations. Temporary databases are used by applications for
which the data is so transient that there is no value in recovering the database after a system failure, for example,
highly dynamic state information in a call center application.
14 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Transaction logging enables the recovery of transactions against persistent databases after a system failure, and
the rollback of transactions.
Locking and Isolation
TimesTen applies locks to prevent a user from changing data that is currently being read or modified by another
user. A lock is placed on database objects at the database, table, or row level. Row-level locking provides the
greatest multiuser concurrency, and is the default.
Applications can call a procedure to change the locking level to row or database during runtime. Table-level locking
is used when the TimesTen optimizer determines that it’s advantageous, or when the application calls a procedure
that directs the optimizer to apply table-level locking for the duration of a transaction.
Users can select an isolation level to specify the behavior of locks that are applied in response to read operations.
The two isolation levels are serializable or read-committed (the default).
Applications select either the serializable or read-committed isolation level. Read-committed isolation provides the
greatest multiuser concurrency. It guarantees that readers will only see committed data values, but doesn’t wait for
writers to release or place locks on records that are being read.
Serializable isolation is used for transactions that demand consistent, predictable data values through the life of a
transaction. With serializable isolation, records that are read are prevented from being updated or deleted by other
users until the transaction commits or is rolled back. During this time, other users aren’t allowed to insert new
records if they would be part of the result set from these read operations. In other words, serializable isolation
guarantees that a read operation could be repeated with the same results. With database-level locking, transactions
are effectively operating with serializable isolation by default.
Read-committed isolation provides the greatest multiuser concurrency. It guarantees that readers will see only
committed data values, but doesn’t wait for writers to release or place locks on records that are being read.
TimesTen creates two copies of a record that is being updated: the preupdated and thus “committed” values for
readers, and the updatable version for the writer. Readers are not blocked access to the read version, and writers
don’t wait for readers.
Query Processing
A primary objective of TimesTen is to adopt relevant open standards. Applications can communicate with TimesTen
databases through SQL, PL/SQL, Pro*C, JDBC, ODBC, TTClasses, JMS, and OCI interfaces.
Most specialty products designed for high performance require proprietary APIs. Oracle TimesTen does not. The
only way to access an Oracle TimesTen database is through standard languages and interfaces and through
Oracle’s popular languages and interfaces. In particular, Oracle TimesTen databases are accessed through: the
structured query language (SQL), the procedural language/structured query language (PL/SQL), Pro*C, Java
database connectivity (JDBC), open database connectivity (ODBC), TTClasses, Java Message Service (JMS), and
the Oracle Call Interface (OCI). The implementation of these interfaces has been highly tuned for the architecture of
TimesTen products.
SQL has been widely adopted for years as the query language standard for relational databases. Abstraction from
the underlying storage and indexing details is one of the main benefits of SQL. It’s also an easy-to-use language
15 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
that expresses which action is to be done, rather than how to perform it. SQL has no references to indexes, data
types, or physical layouts—just tables, columns, and search conditions. The optimizer feature in TimesTen
determines the fastest way to respond to a query based on factors such as the presence of indexes and the
distribution of key values.
This level of abstraction allows the underlying data model to be tuned or extended without affecting existing
applications. New services can be quickly added into a production environment simply by adding application
modules and any required data tables and columns. Without a query language such as SQL shielding the
application from the internals of the data manager, most applications would be affected by the addition of new data
fields.
Data Replication
Replication is configured through SQL statements and can apply to designated tables or an entire database. To
enable high efficiency and low overhead, Oracle TimesTen products use a transaction-log-based replication
scheme.
Replication is the process of copying data between databases. This can help make data continuously available to
mission-critical applications with minimal performance impact. In addition to its role in failure recovery, replication is
also useful for distributing user loads across multiple databases for maximum performance and for facilitating online
upgrades and maintenance.
TimesTen follows a master/subscriber replication model in which committed changes are copied from their source to
one or more subscriber databases. Replication is configured through SQL statements and can apply to designated
tables or an entire database. To enable high efficiency and low overhead, a transaction-log based replication
scheme is used.
Replication at each master and subscriber database is controlled by replication agents that communicate through
TCP/IP stream sockets. The replication agent on the master database reads the records from its transaction log and
forwards any detected changes to the replication agent on the subscriber database. The replication agent on the
subscriber then applies the updates to its local database. If the subscriber agent is not running when the updates
are forwarded by the master, the master retains the updates in its transaction log until they can be applied by the
subscriber.
Balancing Performance and Consistency
The master and subscriber databases have internal mechanisms to confirm that the updates have been successfully
received and committed by the subscriber. These mechanisms, which are completely invisible to the applications,
ensure that updates are not lost and are applied by a subscriber only once.
The replication mechanism in TimesTen is by default asynchronous. When using asynchronous replication, an
application updates a master database and continues working without waiting for the updates to be received by the
subscribers. The master and subscriber databases have internal mechanisms to confirm that the updates have been
successfully received and committed by the subscriber. These mechanisms, which are completely invisible to the
application, ensure that updates are not lost and are applied by a subscriber only once.
16 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Asynchronous replication provides best performance, but the application is decoupled from the receipt process of
the replicated elements on the subscriber. TimesTen also provides two return service options for applications to
confirm that the replicated data is consistent between the master and subscriber databases.
» The return receipt service loosely couples or synchronizes the application with the replication mechanism by
blocking the application until replication confirms that the update has been received by the subscriber
» The return twosafe service enables fully synchronous replication by blocking the application until replication
confirms that the update has been both received and committed by the subscriber
Applications that use the return services trade some performance to ensure higher levels of data integrity and
consistency between the master and subscriber databases.
Replication Topologies
TimesTen supports many replication topologies, but recommends the active standby pair configuration for highest
availability. This is also the only configuration that is supported with the TimesTen Cache.
Figure 5. Active Standby Pair Replication
An active standby pair includes an active master database, a standby master database, and optional read-only
subscriber databases. The active master database is updated directly. The standby master database cannot be
updated directly. It receives the updates from the active master database and propagates the changes to read-only
subscriber databases. This arrangement ensures that the standby master database is always ahead of the read-only
subscriber databases and enables rapid failover to the standby database if the active master database fails.
Only one of the master databases can function as an active master database at a specific time. If the active master
database fails, the role of the standby master database must be changed to active before recovering the failed
database as a standby database. The replication agent must be started on the new standby master database.
If the standby master database fails, the active master database replicates changes directly to the read-only
subscribers. After the standby master database has recovered, it contacts the active standby database to receive
any updates that have been sent to the read-only subscribers while the standby was down or was recovering. When
the active and the standby master databases have been synchronized, then the standby resumes propagating
changes to the subscribers.
17 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Automatic failure detection and failover of database and applications is available through integration with Oracle
Clusterware.
Active standby replication can be used with the Oracle TimesTen Application-Tier Database Cache to achieve
cross-tier high availability. Active standby replication is available for both read-only and updatable caches.
“Leading service providers rely on our products to manage time-critical billing data, so bottlenecks and downtime
cannot be tolerated. With Oracle TimesTen and Oracle Real Application Clusters, we gain a reliable, scalable,
proven real-time data management foundation to better serve our customers.”
Ed McKee,
Director of Applications, Interact, Inc.
Caching
Customized Integration
It’s common for TimesTen to be deployed in conjunction with a disk-based RDBMS, so that TimesTen is used when
data needs to be captured or processed in real time. As the data transitions to a non-real-time state (for example,
when a stock trade is complete or a call detail record has been rated) the information is transferred from TimesTen
to the back-end RDBMS. There are multiple ways in which this integration can be achieved.
Applications can connect to both TimesTen and the back-end database, and move the data through regular API
requests. This is the most flexible approach, but is the least transparent to the applications and could require
complex programming. For example, an application that needs to cache all active subscribers could first request a
record in a TimesTen database, and if results are not found, connect to the back-end RDBMS and repeat the same
request, and insert the results into the TimesTen database. If there were any updates to the data, they would need
to be made in both databases by the application. However, changes made directly to the data in the back-end
RDBMS could result in cache coherency issues.
A variation of this approach is to connect TimesTen and the back-end RDBMS through a publish-and-subscribe
message bus, write new modules separate from existing applications that listen for changes, and then duplicate
changes picked up from the bus. An application can use TimesTen’s transaction log API (XLA) to register for and
receive notice of updates. Most RDBMSs offer a trigger feature that can signal changes, or provide an API similar to
XLA.
Oracle TimesTen Application-Tier Database Cache
A pre-integrated caching solution for the Oracle Database is through the Oracle TimesTen Application-Tier
Database Cache (TimesTen Cache) option. This option enables an application to cache subsets of rows and
columns from Oracle Database tables in the TimesTen Database. Cached data may be pre-loaded by the
application, or may be faulted in the cache on demand. Cache data can be read and updated, and the TimesTen
Cache synchronizes data between the two databases automatically.
A TimesTen cache grid is a collection of TimesTen Caches that collectively manage an application’s cached data. A
cache grid consists of one or more grid members each backed by a TimesTen Cache. Grid members cache tables
from a central Oracle database or Real Application Cluster (RAC). Cached data is distributed across multiple nodes
without shared storage. A cache grid provides applications with location transparency for cached data and ensures
that data is consistent across nodes.
18 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Applications access data via SQL or PL/SQL statements using one of the TimesTen supported APIs, namely ODBC,
JDBC, ADO.NET, OCI, or TTClasses, a proprietary C++ interface.
Event Processing
Event processing applications can sense and respond to events deemed to have business relevance.
There are multiple ways in which Oracle TimesTen enables event-processing applications. At the most basic level,
TimesTen is designed to support applications that must react in real-time to business events. The ability to co-exist
with applications on a wide range of computing servers provides the flexibility to support event-processing
applications.
However, simple events by themselves often do not reveal their relevance and must be correlated with other
information or staged until a pattern is evident. In these cases, the event data must be captured in a data manager
and possibly compared to reference data that was pulled from another data source. When an event that requires
action is identified, taking the appropriate response must, in many cases, be instantaneous. Such processing is
ideally suited to the functionality of TimesTen.
The TimesTen Transaction Log API (XLA) enables the detection of database updates. Applications use XLA to
monitor changes in a TimesTen database and take actions based on those changes. Notification may be desirable
when, for example
» The value of any customer’s total trades for the day exceeds US$1,000,000
» A customer on the “A” list makes a purchase
» A prepaid balance is depleted
» A network element is taken offline
» A flight changes departure gates
Multiple applications can simultaneously read transaction log updates, and each application can maintain its own
bookmark in the log file to maintain its position. Bookmarks are persistent across database connections, shutdowns,
and system failures, so event notifications can pick up where they left off.
XLA is often used to build a custom data replication solution to a non-TimesTen database. TimesTen’s event
processing functionality can be achieved in a conventional RDBMS through triggers and stored procedures.
However, XLA is designed for lower overhead and higher performance, consistent with real-time expectations. Also,
when combined with materialized view functionality, XLA events can be targeted at very granular subsets of the
database. Traditional database triggers execute their logic every time a change is made to any record in the table.
Materialized views combined with XLA functionality provide an efficient way to implement fine-grained event
notification. That is, a materialized view pulls together data that is related to specific events that might have actions
taken on it. An XLA application needs only to monitor update records that are of interest from a single materialized
view table. Without a materialized view, the XLA application would have to monitor all the update records from all
the base tables, including records reflecting updates to rows and columns of no interest to the application.
19 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Conclusion
With the growing speed of messages moving through business networks, the use of real-time processing to capture,
analyze, and respond intelligently to key events is becoming the benchmark for corporate excellence. This isn’t
important simply for the execution and management of critical business processes. Customers expect highly tailored
interactions and the utmost responsiveness from any company with which they do significant business.
What’s needed is a generation of lightweight infrastructure software that presents familiar, powerful interfaces and
query languages that are widely in use—that can easily interface with existing back-office databases, messaging
systems, and application servers—and exploits the full performance potential of today’s networked, memory-rich
computing platforms. This is the generation of infrastructure software for real-time data management provided by the
Oracle TimesTen product.
TimesTen provides application-tier data management for performance-critical systems, optimized for blazing-fast
response and real-time caching of Oracle data. Thousands of companies worldwide, including Alcatel-Lucent,
Aspect, Avaya, Bank of America Merrill Lynch, Bridgewater Systems, BroadSoft, Cisco, Deutsche Börse, Ericsson,
JPMorgan, KDDI, NEC, NYFIX, Smart Communications, United States Postal Office and Verizon Wireless use
Oracle TimesTen in production applications.
20 | EXTREME PERFORMANCE USING ORACLE TIMESTEN IN-MEMORY DATABASE
Oracle Corporation, World Headquarters
Worldwide Inquiries
500 Oracle Parkway
Phone: +1.650.506.7000
Redwood Shores, CA 94065, USA
Fax: +1.650.506.7200
CONNECT W ITH US
blogs.oracle.com/oracle
twitter.com/oracle
Copyright © 2014, Oracle and/or its affiliates. All rights reserved. This document is provided for information purposes only, and the
contents hereof are subject to change without notice. This document is not warranted to be error-free, nor subject to any other
warranties or conditions, whether expressed orally or implied in law, including implied warranties and conditions of merchantability or
fitness for a particular purpose. We specifically disclaim any liability with respect to this document, and no contractual obligations are
formed either directly or indirectly by this document. This document may not be reproduced or transmitted in any form or by any
means, electronic or mechanical, for any purpose, without our prior written permission.
oracle.com
Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective owners.
facebook.com/oracle
Intel and Intel Xeon are trademarks or registered trademarks of Intel Corporation. All SPARC trademarks are used under license and
are trademarks or registered trademarks of SPARC International, Inc. AMD, Opteron, the AMD logo, and the AMD Opteron logo are
trademarks or registered trademarks of Advanced Micro Devices. UNIX is a registered trademark of The Open Group. 0614
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

advertising