Tuning SANs with SSDs - by Craig Harries
Imperial
Technology
(this article was published here in August 2002)
Introduction
No
one will dispute the storage industry is in the midst of a revolution.
The
dynamic growth of the Internet and e-business applications have created new
challenges of how to effectively manage the growing stockpiles of corporate
information, to ensure timely access, and continuous availability.
SANs
Today's
fundamental solution to this problem is the SAN. What is a SAN? A SAN is a
means to allow multiple servers to have direct access to common storage devices
or a storage pool. A SAN can also be viewed as an alternative network dedicated
solely to storage resources.
SANs reduce the burden placed on web,
database, data warehouse, ERP, and CRM applications by consolidating storage
onto its own manageable network that can be administered as a whole rather than
many disparate resources. SANs also provide a better strategy for backup and
managing heterogeneous environments. By placing storage resources on a
dedicated high-speed network, numerous benefits can result:
- - Centralized data eases movement between servers for improved resource
utilization.
- - Manageability is improved as resources can be viewed as a consolidated
pool
- - Efficient backup either LAN-less or Server-less
- - Improved Return On Investment (ROI)
Clearly, Storage Area Networks (SAN) are the storage building
block of the future.
FACTOID: Administrators will be able to support 400% more
storage in a SAN environment than current architectures.
Solid State Disks
Storage devices commonly
deployed in SANs include Cached Disk Arrays, JBOD, Tape Drives, and Solid State
Disks. Solid State Disks play an increasingly important role in SANs as their
scalable design allows them to participate in SAN Islands (standalone SANs) or
in centralized SANs consisting of sub-SANs (multiple SAN Islands interconnected
through Director class products). Because SSDs can be shared by multiple
heterogeneous servers, SAN Island segments, or Clustered nodes, the
opportunities to architect a SAN around the performance and scalability of SSD's
can solve a myriad of performance as well as performance tuning issues within
the SAN fabric. SSD storage resources can be flexibly apportioned to different
SAN needs and managed thru the SAN with standard tools. This manageability
allows the performance benefits of SSD to be effectively directed to areas of
the SAN where performance acceleration is needed. SSDs complement the other
storage resources in the SAN by ensuring that frequently accessed data is
statistically and consistently positioned on the fastest media available.
Even
with all the intrinsic benefits, SANs are not the cure-it-all for all storage
management headaches. While many of the problems associated with server
attached storage are solved by SANs, they may cause a whole new spectrum of
issues. Like any new technology, solving one set of problems tends to highlight
additional issues that need to be understood and acted upon.
Achieving
the ultimate benefits of SAN implementation requires architecting an underlying
virtualization scheme with enterprise-class features, along with a flexible
approach that scales with future needs. The starting point for implementing a
SAN is determining the criticality of the data storage:- how fast is fast
enough? and to what degree will it need to change in the future? The more
substantial the requirements, the more tightly integrated the SAN architecture
must be. This means getting optimal performance, maximizing utilization of the
resources, providing open interfaces to a variety of application needs, and
having the widest range in scaling performance, availability, and reliability.
FACTOID: SANs will enable users to use 90%+ of total storage
resources vs. as little as 40% utilization today.
Utilization and Performance The key to
configuring SANs for high performance database applications is to avoid
contention or bottlenecks. So, when creating a SAN for high transaction-based
database environments, avoid the mistake of trying to use a single Fibre HBA or
loop to support the database. Instead use multiple FC HBAs to spread I/O
devices on different interfaces to avoid contention. For the especially high
transaction files such as redo segments, rollback logs, swap space and other
index types, these should be isolated from the database structures since these
file types can quickly become a performance bottleneck.
Like any
other storage architecture, SANs can benefit immensely from performance tuning.
Since a SAN is significantly more complex than a Cached Disk Array, for example,
tuning a SAN is more complex as well. The key to tuning a SAN is attaining
information on specific areas of performance. Among other things that you need
to know to effectively tune a SAN, are throughput rates by device, and average
daily throughput (rates typically measured in I/O's per second in a database
environment), and average block or record size. Good SAN management software
should provide all this information. The key metrics are block/record size and
I/O's per second.
Typically, a SAN is optimized to move data in
fairly large blocks, which illustrates why Fibre Channel architectures were
initially embraced by video and pre-press operations. Usage in database
environments was slower in coming as the average record size in many databases
tends to be much smaller, not uncommonly 2k blocks, which may not compare
favorably with SCSI connectivity, which is more suited to small block sizes.
This difference emphasizes the need to architect the SAN around these variables
for acceptable performance and scalability.
Another concern in SAN
implementations should be Quality of Service (QOS). Since the SAN
infrastructure shares common components (Switches, Directors, and storage
resources), a valid concern is ensuring that high priority data requests are
promptly responded to and are not queued behind pending low-priority data
requests. This introduces the concept of developing a Hierarchical storage
model within the SAN for QOS purposes. Once again, the Solid State Disk can be
employed for the pinnacle of performance and to ensure a statistically
consistent level of data accessibility and performance.
The rate of
change in SAN components has lead industry analysts like
Gartner to caution SAN
implementers to plan for a three year depreciation schedule for SAN componentry
in part driven by changing standards and in part by the advance of new
interconnect technologies such as
InfiniBand,
iSCSI, and DAFS.
To take one step further, investment protection is a critical factor in
architecting a SAN, and one that should be considered when making suitable
component choices. The ideal SAN component would have a soft interface which
would allow the component to be adaptable over time to new interconnect
technologies and scale appropriately as needs change in the future. Here again,
solid state disks deserve further investigation. Regardless of the interconnect
employed, servers cannot approach the almost limitless capabilities of today's
SSD devices. A server processing 10,000 I/O's per second with a 4kb block size
is only moving 40MB/s of data well beneath the 160MB/s capabilities of a
single channel on an SSD. Given leading SSD's feature 8 to 16 connections, the
longevity of the SAN-based SSD should easily exceed three years as long as its
is adaptable to evolving and changing interconnectivity.
Storage
Virtualization
Virtualizing a solid state disk into the storage
pool opens up some very interesting scenarios. The SSD becomes a strategic
resource that can be directed within the fabric to specific performance issues
and seamlessly migrated between platforms, operating systems, and applications
for true point and click performance resolution. For example, the SSD is
utilized on January 31st to speed-up month-end closing on an NT
based financial application. On February 1st, the SSD (or a portion
thereof) can be allocated to help Engineering with a Unix based OS conversion.
This mobility and scalability allows the organization to effectively leverage
the acquisition across a broader application spectrum and recognize the benefits
in virtually every facet of the enterprise.
Conclusion
SANs
are enabling technologies that are in the adoption phase of the technology
lifecycle. The networked capabilities of the SAN lend themselves to deploying
enabling technologies in an increasingly more powerful and broad-based role
across the organization. The SAN challenge is to understand the data
requirements and architect and tune the SAN for maximum benefit to the
enterprise. Architecting solid state disk into the fabric is a powerful and
easily leveraged resource that can be used and quantified for maximum Return On
Investment. ...Imperial
Technology profile
|
|
. |
|
.... |
About the Author:- Craig
Harries has worked within the technology industry for over twenty years for both
hardware and software companies. He has been very involved with solid state
acceleration products for eight years and is currently vice president of Product
Marketing at Imperial Technology, Inc. | |
. |
|
. |
|
. |
"...You
like the idea - SSDs could make your apps go faster. Problem is - you're
not in an industry where you can stuff raw low latency and high
IOPS in one end of your business sausage machine and expect to see
increased revenue and dollars streaming out the other end..." |
...from:-
the need for auto tiering
SSDs / SSD ASAPs | | |