Optimizing MS Exchange with the
QLogic 10000 Series Adapter
FabricCache Technology Processes Exchange
Transactions Up to Five Times Faster
FabricCache technology makes caching I/O data from
Exchange entirely transparent to the host, which improves
reliability and reduces complexity.
White Paper
KEY FINDINGS
• The performance of Microsoft® Exchange Server® is driven by the IOPS
capability of the underlying storage solution. The QLogic® FabricCache™
10000 Series 8Gb Fibre Channel Adapter multiplies the capabilities
of existing storage technology by integrating Fibre Channel storage
network connectivity, ash caching, and embedded processing. QLogic
10000 Series’ FabricCache technology makes caching I/O data from
Exchange entirely transparent to the host, which improves reliability and
reduces complexity.
MICROSOFT EXCHANGE EVOLUTION AND ARCHITECTURE
The beating heart of the Exchange Server is the Microsoft Extensible
Storage Engine (ESE) API, also known as Joint Engine Technology (JET)
Blue. ESE is an indexed sequential access method (ISAM) data storage
technology that also makes up the core of Microsoft Active Directory®. ESE
provides transacted data update and retrieval capabilities, as well as a
crash recovery mechanism to maintain data consistency even in the event
of a system crash. Transactions in ESE are highly concurrent, making ESE
suitable for server applications such as Exchange.
The I/O Gap
ESE and Exchange have beneted greatly as processor and memory
performance keep pace with the promises of Moore’s Law. Storage
technology has unfortunately lagged well behind the trend. This gap leaves
Exchange starved for IOPS, which is the most critical factor in transactional
performance. Microsoft heralded several new design features to help cope
with this IOPS shortfall. Among them is the advent of database availability
EXECUTIVE SUMMARY
Few applications describe “mission critical” like the enterprise e-mail
engine. An estimated 918million users start their day by checking
e-mail messages, appointment calendars, and task lists. The alwayson, always-connected global worker is never too far from this vital
link.
The increased demands on the Microsoft Exchange Server make
it critical to optimize the server’s I/O performance in the SAN. The
QLogic 10000 Series 8Gb Fibre Channel Adapter meets that need for
high performance. By integrating a unique I/O caching architecture
with Fibre Channel connectivity, the 10000 Series Adapter
signicantly increases the performance of transactional I/O (IOPS)
and I/O read latency for large Exchange data volumes. This combined
approach of enterprise server I/O with server-based I/O caching using
ash memory delivers dramatic and smoothly scalable application
performance.
This document is intended for IT leadership, Exchange administrators,
and all those accountable for the performance of the Exchange
messaging infrastructure.
SN0430948-00 Rev.B 11/13 1
Optimizing MS Exchange with the
QLogic 10000 Series Adapter
White Paper
groups (DAGs) to disburse the I/O load, as well as key changes to the
Exchange caching mechanism. The most notable change is a write caching
feature that collects the pieces of data headed for storage and organizes
them in sequential, contiguous blocks. Contiguous blocks require fewer
IOPS to retrieve the data. Therefore, data read operations see an indirect
boost from this front-end operation, as shown in Figure1.
Figure1. Bridging the I/O Gap
Improving Transactional I/O with QLogic 10000 Series
Despite these new improvements, Exchange transactions are still
essentially small-block, random I/O operations that produce heavy loads
on the storage array. The QLogic 10000 Series Adapter is designed
especially to accelerate this specic kind of workload. The 10000 Series
Adapter works seamlessly with Exchange to accelerate the delivery of
e-mail.
EXCHANGE STORAGE DESIGN
Storage design is one of the most important elements of a successful
deployment of Microsoft Exchange Server. To achieve a storage design
that is optimized for reliability, performance, and price, it is critical that you
follow specic, recommended guidelines.
Performance First
Two sizing characteristics dene the storage needs of Exchange: capacity
and performance. Capacity is the raw storage needed to hold all of the
messages, attachments, and meeting notes. Performance is the ability to
move that data to and from storage. While both features are important,
storage design for Exchange solutions should rst consider performance
because it is the most challenging quality to control. Performance is driven
by the IOPS capability of the storage solution.
Exchange 2010 Mailbox IOPS Prole
A key ingredient to a successful deployment of Exchange is an accurate
assessment of individual usage patterns. These patterns are described in a
mailbox prole.
Table1 provides estimated values for IOPS per user that you can use to
predict your baseline Exchange 2010 IOPS requirements and includes all
database I/O (database, content indexing, and NTFS metadata). It does not
include log volume I/O, which typically adds 10percent to the workload.
Messages
Sent and
Received
per
Mailbox
per Day
5030.061:10.053:2
10060.121:10.103:2
15090.181:10.153:2
200120.241:10.203:2
250150.301:10.253:2
300180.362:30.301:1
350210.422:30.351:1
400240.482:30.401:1
450270.542:30.451:1
500300.602:30.501:1
Table 1. Database Cache and Estimated IOPS Per Mailbox Based on Message Activity
Database
Cache per
Mailbox
(MB)
Per Mailbox with a
Single Database Copy
Estimated
IOPS
Read/
Write
Ratio
Per Mailbox with
Multiple Database
Copies
Estimated
IOPS
Read/
Write
Ratio
After a baseline is established, other tools from Microsoft can further rene
the expected IOPS load:
• The Microsoft Exchange Server Prole Analyzer collects statistical
information about existing Exchange environments to provide data to
dene user proles.
• The Exchange 2010 Mailbox Server Role Requirements Calculator helps
to quantify the needs for new and planned environments.
STORAGE DESIGN AND POTENTIAL IOPS
Every component in the storage path has an impact on throughput.
Ultimately, drive speed and type dene the maximum IOPS potential. Faster
drives support more IOPS, as depicted in Figure2.
The two most signicant factors that predict Exchange 2010 database
IOPS are the amount of database cache provided per user and the quantity
of messages each user sends and receives per day. Table 1, published
by Microsoft, is based on a standard worker who uses Outlook 2010 in
Cached Exchange Mode. The information has been tested to be accurate
within plus or minus 20percent using database cache sizes between
3MB and 30MB per mailbox. The average message size for validation was
75KB, although message size is not a primary factor for IOPS.
Figure 2. Drive Speed Affects E-Mail Performance
SN0430948-00 Rev.B 11/13 2
Optimizing MS Exchange with the
QLogic 10000 Series Adapter
White Paper
Hard drives can be combined into RAID sets to increase the IOPS potential,
as shown in Figure3. RAID sets also increase the presented capacity,
which allows for larger databases.
In addition to greater IOPS potential and capacity, RAID sets offer various
degrees of high availability. A database will continue to be available even
after a physical drive failure. The cost of RAID is a specic IOPS penalty
that occurs during the write cycles that accounts for the redundancy
provided. Table2 shows the write penalty for common RAID types.
RAID TypeStructure
RAID 1
RAID 10
RAID 5
RAID 6
1
N = quantity of disks; C = capacity of each disk.
Equal number of data
disks and mirrored disks
Equal number of striped
data disks and mirrored
disks
Two or more data disks;
one striped disk for parity
Two or more data disks;
two striped disks for
parity
Table 2. Write Penalty for Common RAID Types
Concatenated
Capacity
(N × C) / 22
N / 22
N × C – (C × 1)4
N × C – (C × 2)6
1
Write IOPS
Penalty
The formulas used to calculate the potential IOPS of a solution using RAID
are as follows:
Specic Example from an IOPS Perspective
A storage solution is designed for 5,000users who send and receive
100 messages a day (5,000users × .120 = 600IOPS). QLogic highly
recommends adding an additional 20percent margin for unexpected
spikes in activity. This calculation produces an IOPS target workload
of720.
This example uses RAID5 to guard against disk failure, which incurs a
write penalty of4. Deploying multiple database copies produces a read/
write ratio of 3:2 or 60percent writes and 40percent reads, as shown in
Table3.
Write
Disk
Speed
(RPM)
7.2k85161360204544748
10k125111375206550756
15k17581400210560770
1
Raw IOPS × .6 / 4
2
Raw IOPS × .4
IOPS
per
Disk
Quantity
of Disks
Needed
Raw
IOPS
Table 3. IOPS Example
IOPS
with
RAID 5
Penalty
Read
IOPS
1
2
Solution
IOPS
Note that the quantity of disks required to support this workload increases
substantially for slower disks. This increase is contrary to the thinking
of using slower-but-larger disks to meet capacity requirements. Larger
capacity serial advanced technology attachment (SATA) hard drives
offer an appealing price per gigabyte (GB), but do not have strong IOPS
capability. Higher-performing hard disks such as modern serial attached
SCSI (SAS) drives are more expensive per GB, but can better withstand the
rigors of the IOPS workload of Exchange.
FABRICCACHE OFFLOADS IOPS OVERHEAD
QLogic’s patent-pending FabricCache technology is an innovative
approach to releasing Microsoft Exchange Server from its heavy IOPS
burden. The QLogic 10000 Series Adapter resides within the mailbox
server and transparently caches the ow of information to and from the
Exchange database. Transparency means no special conguration or
tuning of Exchange Server is required. The ash-based cache captures the
most frequently used Exchange information and presents it at speeds often
ve times faster than a regular call to storage, as shown in Figure4. With
the IOPS load under control, storage designers now have greater exibility
in the speed, size, and cost of their drive conguration.
Note: Only write operations incur a penalty for RAID.
SN0430948-00 Rev.B 11/13 3
Optimizing MS Exchange with the
QLogic 10000 Series Adapter
White Paper
SUMMARY
Microsoft Exchange is a mission-critical, highly utilized, storage-hungry
application. Proper storage design is vital to a successful Exchange
deployment. Providing adequate IOPS capability is an important
consideration in the storage design formula.
Higher-performing disks offer better IOPS capability. When coupled with
the QLogic 10000 Series FabricCache Adapter, the disks process Exchange
transactions as much as ve times faster. This combined solution offers
transparent integration, seamless interoperability, and rock-solid reliability.
Figure 4. QLogic 10000 Series Adapter Reduces IOPS Load on Storage Array
RESTORE “LIKE-NEW” RESPONSE
Existing Exchange solutions that utilize Fibre Channel technology will
For more information about how QLogic solutions speed up processing
with Exchange, visit QLogic’s Microsoft Exchange Server Web page:
benet greatly from an upgrade to the QLogic FabricCache 10000 Series
Adapter. The Host Bus Adapter upgrade is a simple process that requires
no modication to the Exchange environment or the storage platform. The
performance gain is immediate and truly remarkable. The upgrade can
breathe new life into your Exchange solution, restoring that “like-new”
responsiveness. The performance enhancement can extend the useful life
of your storage solution by supporting more transactions by more users.
DISCLAIMER
Reasonable efforts have been made to ensure the validity and accuracy of these performance tests. QLogic Corporation is not liable for any error in this
published white paper or the results thereof. Variation in results may be a result of change in conguration or in the environment. QLogic specically
disclaims any warranty, expressed or implied, relating to the test results and their accuracy, analysis, completeness or quality.