Automated Data-Aware Tiering

Automated Data-Aware Tiering
White Paper
Drobo’s revolutionary new breakthrough technology automates the
provisioning, deployment, and performance acceleration for a fast
tier of SSD storage in the Drobo B1200i array. Customers can now
take advantage of traditional HDD capacity plus SSD acceleration
across all of their applications automatically and immediately.
Introduction
What Is Data Aware?
Drobos are data aware, that is,
they intelligently handle data for
added resiliency, performance, and
management. Drobo is data aware in
two areas. First the Drobo knows the
difference between data and nondata; traditional RAID does not. This
allows Drobo to provide benefits that
are unique in the industry including
an array-level capacity gauge, virtual
hot-spare, proportional rebuild times,
automatic space reclamation for
deleted data, and much more.
Second, Drobo is aware of the
difference between streaming and
transactional data, which enables
instantaneous and intelligent tiering of
data. As data is written to the Drobo,
it is analyzed in flight and intelligently
placed on the correct media (SSD
or HDD) to maximize performance.
This increases the benefit of SSDs
by eliminating the lag that can occur
with other tiering solutions that must
analyze I/O patterns before placing
data on the appropriate tier.
Storage has traditionally been difficult to understand and manage, especially for non-storage
experts. Tasks such as pooling drives, scaling capacity, and protecting from data loss are
complex tasks for individuals and businesses that require high-capacity and reliable storage.
Also, the explosion of data at every level of business requires more sophisticated digital
storage. The simple fact is that legacy solutions are typically too complex or too limited for
current needs.
Unique Drobo technology delivers breakthrough ease-of-use, affordability, and value so that
individual professionals and businesses can have the best storage experience ever. The first
Drobo innovation, known as BeyondRAID™, provides a unique combination of simplicity and
storage sophistication. BeyondRAID breaks down the barriers of traditional RAID in order
to greatly simplify storage deployment and management and to protect data from a drive
failure. Automated Data-Aware Tiering works seamlessly with BeyondRAID to optimize data
placement automatically based on the type (or “tier”) of data being stored.
SSDs Enable Automated Data-Aware Tiering
SSD (solid state drive) support in the Drobo B1200i creates a faster tier of storage for data that
is transactional in nature, the type of data that benefits from the faster read and write speeds
that SSDs provide. But how do small and medium business (SMB) IT administrators effectively
use SSDs to accelerate their applications? They can’t. Ether the array is too expensive (costing
more than $50,000) or, even if the SMB-focused arrays allow SSDs to be used, they must be in
their own pool and separate from the rest of the data. Data must then be manually configured
to be sent to the fast volume (SSD) or to the normal volume (hard disk drives, or HDDs).
Physically separated storage tiers force administrators to segregate their traffic and direct it
to different types of storage, which is both challenging and time consuming. But how can IT
administrators figure out what data is transactional in nature, that is, what data should be
sent to the fast volume? Often data has mixed characteristics, making it even more confusing
and complicated. To get the most out of expensive SSDs, the solution must be intelligent and
fully automatic—data just needs to go to the right place automatically.
Figure 1:
Data is analyzed in-flight as it enters a Drobo
and is placed intelligently on the drives.
Automated Data-Aware Tieringdrobo.com |
2
A Drobo is a single pool of drives by design. Users do not need to think about pooling drives or
creating RAID groups. Also, drives of different sizes and types can be used in a Drobo, making
Drobo ideal for helping SSDs and HDDs work together optimally and automatically in the same
chassis.
To support drives of different sizes, the Drobo virtualization layer sits on top of the physical
drives (find out more about Drobo BeyondRAID at www.drobo.com/how-it-works/beyondraid).
Because Drobos are data aware, data is placed intelligently in the best place on the drives.
Being Data Aware
As data enters the Drobo—even while it is still in flight—it is already being examined, but
note that performance is not impacted. An additional check determines whether or not the
data is transactional in nature. If it is transactional, then the data is written to the faster
SSD tier.
Figure 2:
As data is written to the Drobo, transactional data (shown in red) is written to the
transactional tier (SSDs) and the remaining
data is written to the bulk tier (HDDs).
Even leading enterprise storage arrays that offer tiering capabilities and typically cost more
than $50,000, are not always aware of what type of data being stored. Often data is first
written to bulk storage and then analyzed before it is sent to a high-performance tier—and
this takes time. With Drobo, transactional data goes directly to SSDs immediately, which
ensures that reads AND writes are optimized.
Drobo’s tiering functionality allows you to store any type of data on the Drobo and the type
of data that can by optimized by SSDs is automatically optimized in flight. This means that all
data—mixed data types from VMs in a cluster, Exchange databases/datastores, backup data,
or files on a file server—automatically goes to the right place.
Streaming vs. Transactional Performance
When streaming data to and from storage (that is, writing and reading files or storing backup data) the data set is large and
operations are often sequential. When small bits of data—such as metadata or fields in a database—are sent to and from storage
operations are often random. Therefore, the data and access patterns of streaming and transactional performance require different
metrics from the storage array.
• When streaming large amounts of data, the metric MB/s (number of megabytes per second that can be written to or read from
storage) is the most important.
• When storing data that is transactional in nature, the metric IOPS (number of inputs and outputs per second that can be written
to or read from storage) is the most important.
When you are evaluating storage to ensure that it can meet the needs of your applications, you need to take into account both
metrics—MB/s and IOPS. The most intelligent way to have the best of both worlds in business-critical environments is to use
storage that has automated tiering capabilities.
Automated Data-Aware Tieringdrobo.com |
3
One Pool, Multiple Tiers
BeyondRAID technology allows the Drobo B1200i to behave as a single pool of storage.
Whether you put 3 drives,12 drives, or any number in between, they are all managed as a
single pool in the Drobo.
Figure 3:
Two tiers of storage automatically configured
in the same pool of storage on the Drobo.
If you have 8 x HDDs and 2 x SSDs or 9 x HDDs and 3 x SSDs, they will all be utilized
appropriately without any preliminary configuration. All the administrator has to do is put the
drives in the Drobo.
Fully Automated Tiering
After data is written to storage, data patterns often change. Sometimes a large amount of
data is written to storage and not read for weeks or even months. And it is not uncommon
for a small amount of data to be written and accessed very frequently. Traditional storage
arrays treat all data the same way, so if the storage is optimized for streaming performance,
then it will most likely not be optimized for transactional performance (see “Streaming vs.
Transactional Performance”).
With Drobo automated tiering, optimization of data does not stop when the data reaches a
transactional or bulk storage tier. As the data stored on the Drobo is being read, it is analyzed
to look for patterns over time. If data on the bulk tier is frequently read and begins to look more
like transactional data, it will be migrated to the transactional tier. If data on the transactional
tier becomes “cold” with very few or no read requests, it will be migrated to the bulk tier.
These migrations occur in the background when the storage is not under high load and do not
require any administrative interaction.
Figure 4:
When data gets “cold” it is migrated to the bulk tier.
When data becomes “hot,” it is migrated to the transactional tier.
Automated Data-Aware Tieringdrobo.com |
Tiering Without SSDs
Even without SSDs, Drobo still puts its automatic tiering feature to work. Without SSDs,
the advantage of tiering is to eliminate the write penalty when parity data is created. One of
the characteristics of BeyondRAID is that a Drobo can automatically have zones of different
types. When all HDDs in the Drobo are the same, Drobo lays out zones differently on the
transactional tier than it does on the bulk tier.
Choosing the Right SSD
Not all drives are the same and if
you do not choose the right one,
performance may not meet your
expectations.
• Although not true in the past,
enterprise-grade drives are now
available that use MLC flash,
significantly driving down the per
GB cost of the drive.
• Enterprise-grade SSDs often have
a higher level of overprovisioned
space, increasing performance and
extending the life of the SSD.
• Enterprise drives are tuned for
extended use to handle more
write cycles and use reclamation
features such as TRIM and
UNMAP to optimize capacity and
performance.
• Similar to HDDs, SSD drives
are available with SATA or SAS
connectivity. Only enterprise-grade
SSDs are available with SAS
connectivity . (The Drobo B1200i
supports both SATA and SAS
drives.)
For these reasons, use enterprise-grade
SSDs and HDDs in your Drobo B1200i
for best performance.
Figure 5:
Drobo creates different zone types on transactional and bulk tiers
to tune the storage appropriately for hot and cold data.
For example, mirror zones are optimized for writes and stripes are optimized for capacity.
Leveraging BeyondRAID technology, zones are created automatically and altered on the fly
so that the Drobo is always optimized for the type of data stored on it.
Just Add SSDs
If your disk pack contains only HDDs, they will be optimized automatically. If your applications
need even more performance, all you have to do is introduce at least two SSDs (for single-disk
redundancy) or three SSDs (for dual-disk redundancy) to the Drobo B1200i. The SSDs then
become the new transactional tier and all of the HDDs are delegated to the bulk tier. This
migration occurs automatically, so the only administrative task is to insert new drives into t
he Drobo.
NOTE: Introducing SSDs into a Drobo increases both write and read performance because
SSDs eliminate drive seek time. This performance boost is critical in database and email
environments, which are characterized by frequent read-write requests that are small in
nature.
Summary
Drobo was created to solve challenges inherent in storage technologies in SMB and enterprise
workgroup environments. Those who were previously familiar with storage assumed that
RAID, which allows drives to be aggregated and protected, was the norm. But it is no longer
the norm. BeyondRAID is a huge advance for users who do not want to have to learn about
storage in order to protect their data and for storage administrators who desire to manage
storage more efficiently.
Automated Data-Aware Tiering takes storage automation to the next level, allowing Drobo
to automate the once-complex tasks of data protection, capacity planning, and performance
tuning (tiering with SSDs).
2460 North First Street, Suite 100, San Jose, CA 95131 • www.drobo.com • 1.866.97.DROBO
Copyright © 2012 • All rights reserved.
Drobo, DroboElite, Data Robotics, DroboPro, DroboStore, BeyondRAID, and Smart Volumes are trademarks of Drobo, which may be registered in some jurisdictions.
All other trademarks used are owned by their respective owners. Specifications subject to change without notice.
WP--0109-00 • March 2012
Click below to find more
Mipaper at www.lcis.com.tw
Mipaper at www.lcis.com.tw

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