Professional Documents
Culture Documents
5371 - Best Practices For Microsoft Exchange 2007 With HP Servers & Storage in Mid-Range Environments
5371 - Best Practices For Microsoft Exchange 2007 With HP Servers & Storage in Mid-Range Environments
MSA1500)
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Solution configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Configuration Tools . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Infrastructure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Test configurations . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Test procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Processor utilization . . . . . . . . . . . . . . . . . . . . . . . . 13
Memory utilization . . . . . . . . . . . . . . . . . . . . . . . . 14
Exchange performance . . . . . . . . . . . . . . . . . . . . . . 23
Testing summary . . . . . . . . . . . . . . . . . . . . . . . . . 26
Exchange administration . . . . . . . . . . . . . . . . . . . . . . . . 28
Storage administration . . . . . . . . . . . . . . . . . . . . . . . . 29
Server administration . . . . . . . . . . . . . . . . . . . . . . . . . 29
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
64-bit platform . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Role-based servers . . . . . . . . . . . . . . . . . . . . . . . . 39
Continuous replication . . . . . . . . . . . . . . . . . . . . . . . 40
Appendix E. Software . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Software updates . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
Overview
In November 2006 Microsoft introduced Exchange Server 2007, and since the release of Service
Pack 1 (SP1), customers have been steadily planning and performing migrations. According to a
corporate IT survey published in Messaging and Collaboration, 2007-2008, September 2007, by
The Radicati Group, Inc., e-mail system upgrades are one of the top five messaging priorities in 2008
for companies in the medium business segment (101-1,000 mailboxes).
Companies want simple, affordable, industry-leading hardware to support their new Exchange Server
2007 environments. The HP StorageWorks Modular Smart Array (MSA) family of storage arrays
combined with HP ProLiant servers provides these requirements. The Customer Focused Testing (CFT)
group has conducted tests in simulated environments of 1,000 users with 750-MB mailboxes on the
MSA60, MSA70, and MSA1500 solutions. The tests compared drive type options within the arrays,
including performance differences, advantages, and disadvantages.
This paper provides configuration and performance data, best practices and recommendations to
help system administrators deploy Microsoft Exchange Server 2007 on several of the HP MSA family
of products. The results presented here are derived from our extensive testing. They are intended
to help you perform pre-deployment planning and ensure adequate hardware and predictable
performance for your specific business requirements.
3
Solution configuration
The Exchange 2007 testing environment included a single, consolidated Exchange 2007
Server with a Mailbox Server, Hub Transport, and installed Client Access Server roles. See
“Microsoft Exchange Server 2007 ” on page 39. The Exchange Server was hosted on an HP ProLiant
DL 380 G5 Server and was evaluated in three different Direct Attached Storage (DAS) solutions.
Using a three-phased testing approach, we configured the Exchange Server with an HP StorageWorks
MSA60, then with an MSA70, and finally with an MSA1500 storage device. We monitored the
performance of the MSA storage devices using the drive types specified in Table 1.
Testing Phase Storage Device Storage Size and Speed Array Controller
Configuration Tools
For each solution configuration, we used the Microsoft Exchange Load Generator (LoadGen) to
monitor Exchange 2007 Server performance under a simulated workload. The LoadGen tool
placed a simulated load of 1,000 users with 750-MB mailboxes using a heavy user profile. See
“Microsoft Exchange Load Generator” on page 42.
The use of HP Storage Planning Calculator (SPC) for Exchange 2007 ensured consistent analysis
for storage sizing and the number of required drives for each MSA and drive type configuration.
In our tests we used both RAID 10 and RAID 5 configurations on database LUNs, again based
upon the SPC output. See “HP Storage Planning Calculator” on page 33. Tests of the various
MSA devices and drive types provided opportunities to validate the SPC results by evaluating and
comparing actual Exchange 2007 performance.
We used the Microsoft Exchange Jetstress performance tool to analyze the optimum disk I/O
capabilities achieved by each configuration.
Infrastructure
Each of the three independent test environments had an identical Active Directory (AD) and network
infrastructure. Each environment consisted of one Windows 2003 Server Domain Controller that
established the Serial Attached SCSI (SAS), Serial Advanced Technology Attachment (SATA), and
Small Computer System Interface (SCSI) test domains. All three domains were child domains in the
same Active Directory Forest, operating at the Windows 2003 functional level. Each storage device
was evaluated under the same workload and stress level in its respective domain for the tested drive
configuration. The major hardware components for each test environment are described in the
sections that follow. For a complete list of components, see “Bill of materials” on page 32.
4
Exchange 2007 Server
The Exchange 2007 Server was hosted on an HP ProLiant DL380 G5 Server equipped with two
dual-core Intel Xeon 5110 (1.60 GHz, 1066 FSB) processors, 16 GB of RAM (PC2-5300 2 X 4 GB),
Microsoft Windows Server 2003 R2 Enterprise x64 Edition operating system with all appropriate
patches and hot fixes, and Microsoft Exchange Server 2007 release to manufacturing (RTM) with
Client Access, Hub Transport, and Mailbox Server roles. For a detailed list of software updates,
see “Software ” on page 46.
The Exchange 2007 Server processor and memory configurations were recommended by Microsoft
for Exchange 2007 multi-role servers (4 x processor cores), 8 GB of server RAM, and an additional
2 MB to 5 MB of RAM for each mailbox.
We followed HP SPC recommendations to connect the Exchange Server to the appropriate storage
device with a directly attached P800 Smart Array Controller (MSA60/70) or HP FC2142SR HBA
(MSA1500). The Exchange Server used two HP NC373i Multifunction Gigabit Server Adapters in a
network-teaming configuration attached to an HP ProCurve 2724 10/100/1000 Ethernet switch.
The Active Directory forest had three individual child domains to host each test environment. Each
domain was autonomous and fully independent. The Windows 2003 Active Directory Servers were
configured on HP ProLiant DL380 G4 Server with two dual-core 3.40-GHz Intel Xeon processors, 4
Gb of RAM, Windows Server 2003 Enterprise Edition OS, and two HP NC7782 Gigabit Server
Adapters.
The Windows AD site was configured as the default site and contained all AD domain controllers.
AD-integrated DNS was installed on the domain controller, which used two HP NC7782 Gigabit
Server Adapters in a network teaming configuration and was attached to an HP ProCurve 2724
10/100/1000 Ethernet switch operating at 1-GB network speed.
The Microsoft Exchange Load Generator placed a simulated workload of 1,000 users with 750-MB
mailboxes. All users were configured with a heavy profile resulting in an estimated .33 IOPS
(Input/Output operations per second) per user. Based on the expected workload, the SPC calculated
that the Exchange 2007 Server would support 12 storage groups and 12 databases. The Exchange
Load Generator clients were hosted on HP ProLiant DL320 servers with Xeon Pentium 4 2.66-GHz
processors, 640 MB of RAM, and two HP NC7760 Gigabit Server Adapters in a network teaming
configuration.
The MSA1500 was configured with enabled Active/Active functionality and connected to two HP
StorageWorks 30 Modular Smart Array SCSI enclosures with 14 X 146-GB 15K SCSI hard drives, one
HP StorageWorks 20 Modular Smart Array with 12 X 500-GB 7.2K SATA hard drives. The MSA1500
was directly connected to the Exchange 2007 Server via two 2-GB FC HBA interfaces.
5
HP StorageWorks 60 Modular Smart Array (MSA60)
We evaluated the MSA60 with 300-GB 15K SAS 3.5-inch hard drives and 500-GB 7.2K SATA
3.5-inch hard drives. The MSA60 was directly connected to the Exchange 2007 Server’s P800 Smart
Array Controller using 1-meter external SAS cables.
We evaluated the MSA70 with 146-GB 10K SAS 2.5-inch hard drives and 120-GB 5.4K SATA
2.5-inch hard drives. The MSA70 was directly connected to the Exchange 2007 Server’s P800 Smart
Array Controller using one-meter external SAS cables.
Test configurations
The three test environments had identical Active Directory and network infrastructure configurations.
The storage device and drive configurations were set up as described in this section and shown in
the configuration diagrams.
Figure 1 shows the Exchange 2007 Server in the MSA70 test configuration, including one Active
Directory domain controller and two Exchange Load Generator clients. The Exchange 2007 Server
was connected to the MSA70 in a DAS configuration. We evaluated the MSA70 with 146-GB 10K
2.5-inch SAS hard drives and 120-GB 5.4K 2.5-inch SATA hard drives (120-GB SATA configuration
required 2 X MSA70 arrays). All servers were connected to an HP ProCurve 2724 10/100/1000
Ethernet switch. The Exchange 2007 Server and the Exchange Load Generator clients incorporated
network teaming.
Figure 2 shows the Exchange 2007 Server in the MSA60 test configuration, including one Active
Directory domain controller and two Exchange Load Generator clients. We connected the Exchange
2007 Server to the MSA60 in a DAS configuration. The MSA60 was evaluated with 300-GB 15K
3.5-inch SAS hard drives and 500-GB 7.2K 3.5-inch SATA hard drives. All servers were connected to
6
an HP ProCurve 2724 10/100/1000 Ethernet switch. The Exchange 2007 Server and the Exchange
Load Generator clients incorporated network teaming.
Figure 3 shows the Exchange 2007 Server in the MSA1500 test configuration, including one Active
Directory domain controller and two Exchange Load Generator clients. The Exchange 2007 Server
was connected to the MSA1500 in a DAS configuration. The MSA1500 was connected to two
MSA30 arrays and one MSA20 expansion enclosure. We evaluated the MSA1500 with 146-GB 15K
SCSI hard drives and 500-GB 7.2K 3.5-inch SATA hard drives. All servers were connected to an HP
ProCurve 2724 10/100/1000 Ethernet switch. The Exchange 2007 Server and the Exchange Load
Generator clients incorporated network teaming.
We configured the physical disks for the Exchange 2007 storage requirements using the HP Array
Configuration Utility. To optimize performance, the disks were separated into two independent
arrays. One array was dedicated to the Exchange 2007 databases, and the other was dedicated to
the Exchange log files. The actual number of drives in each array was dependent on results from
the HP Storage Planning Calculator.
Figure 4 shows the physical disk layout in the Array Configuration Utility of the Exchange database
and log file arrays. Array C was configured to support the Exchange databases. The database
disk array was configured to support 12 x 112-GB logical drives, each dedicated to an Exchange
database. Array D was dedicated to the Exchange log files, and supported 12 x 39-GB logical
drives, each drive supporting one Exchange storage group. The number of drives in an array varied
depending on the HP Storage Planning Calculator results. The number and size of the logical disks
were constant in all test configurations.
8
Figure 4. Array configurations
The Exchange storage group directory structure used volume mount points to present the logical disks
to the operating system, eliminating the need to use drive letters and simplifying administration and
scripting by providing a logical directory structure. Figure 5 shows the Exchange storage group
structure. Each storage group had a dedicated folder in the main Exchange folder on the E: drive.
Each storage group folder also served as the volume mount point for the database and log file disks.
9
Figure 5. Directory structure
10
Testing
Objectives
In our performance tests we evaluated the efficiency of the MSA60, MSA70, and MSA1500
using various SAS, SATA and SCSI drive configurations. The configurations included a ProLiant
DL380GB Exchange Server Solution to host a mid-range Exchange 2007 environment. In each
test we used the same Exchange 2007 user configuration and identical LoadGen workloads. See
“Microsoft Exchange Load Generator” on page 42. The tests identified and compared the impacts
of implementing each drive technology and storage device exposed to the same Exchange 2007
environmental factors.
All three test environments used identical server configurations, Active Directory, and network
infrastructure. The changes made to the test environments involved the storage devices, drive type,
and drive configuration.
Test procedure
1. Plan Exchange storage requirements and the storage group configuration. The HP Storage
Planning Calculator determined the proper Exchange 2007 storage group requirements and the
number and type of hard drives. See “HP Storage Planning Calculator” on page 33.
4. Install server operating systems using the HP SmartStart CD. Install all appropriate patches and
hot fixes using the Windows Update service. See “Software ” on page 46.
5. Install and configure Active Directory on the domain controller and establish the appropriate
test domain.
6. Configure and partition storage using the HP Array Configuration Utility and then use the
Diskpart.exe utility to align disks.
7. Use the Exchange Jetstress utility to analyze disk configuration performance capability (a
6-hour test run).
8. Install Microsoft Exchange 2007 RTM on the Exchange Server. The Exchange Server was
configured with the HUB Transport, Client Access, and Mailbox Server roles.
9. Create and configure the Exchange 2007 storage groups (12) and databases (12) as based on
the results of the HP Storage Planning Calculator.
10. Use the Exchange Best Practice Analyzer to verify healthy Exchange installation.
Note
The following steps were necessary for the testing environment, but are not necessary for production deployments.
11. Install and configure Exchange Load Generator on the Exchange Load Generator clients.
11
12. Use the Exchange Load Generator to create 1,000 test users in the test domain.
Edit the Load Generator configuration XML file to achieve the appropriate Mailbox
size (750 MB) and user workload action profile to match a heavy user profile. See
“Microsoft Exchange Load Generator” on page 42.
13. Use the Exchange Load Generator to initialize each Exchange database to prepare for a
performance test run.
We ran a minimum of three performance tests and one Jetstress test for each storage device and
drive configuration. The Jetstress test validated disk I/O performance capabilities. By running
three tests, we ensured consistent performance results for comparison between configurations. We
used the following procedure:
2. Configure Windows Performance Monitor to capture desired performance counter data for an
8-hour time period.
3. After the initialization/reinitialization process, use the Exchange Load Generator to conduct
an 8-hour performance test.
5. After an 8-hour performance test, reinitialize if necessary, or reconfigure storage device with
the next drive configuration.
• Mean average - An arithmetic mean (or simply the mean) of a list of numbers is the sum of the
list divided by the number of items in the list.
• Standard deviation - The most common measure of statistical dispersion, standard deviation
measures data set value variations. If many data points are close to the mean, the standard
deviation is small; if many data points are far from the mean, the standard deviation is large.
• 95th percentile - This calculation indicates that 95% of the time a performance counter object is
at or below this amount. The top 5% of the performance utilization are short spikes of atypical
usage, and these inflated statistics are discarded. By examining the 95th percentile, we provide a
more accurate representation of the performance counter high points. The 95th percentile is not
an average, but a representation of the highest value reached.
These calculations are based on the Physical Disk/Transfers/Sec performance counter of the
Exchange database array divided by the number of users. The values reflect both the average and
the 95th percentiles of transfers per second and calculated user IOPS. The 95th percentile of user
IOPS shows a maximum workload profile of .2878 IOPS per user by the Exchange Load Generator
clients. A variation of only +/– .0094 exists between each configuration. This data indicates a
12
consistent generated workload by the Exchange Load Generator clients, resulting in consistent and
accurate test performance.
Table 2. User IOPS in RAID 10 drive configurations
Table 3 shows the amount of user I/O achieved by the Exchange Load Generator clients on the
Exchange 2007 Mailbox Server with each storage device and drive in a RAID 5 configuration.
The values show both the average and the 95th percentiles of transfers per second and the calculated
user IOPS. The 95th percentile of user IOPS shows a user maximum workload profile of .2915 by
the Exchange Load Generator clients. A variation of only +/--- .0128 existed between each test
configuration. This data shows a consistent generated workload by the Exchange Load Generator
clients, resulting in consistent and accurate performance of each test configuration.
By comparing the RAID 5 with the RAID 10 results (based on the LoadGen workload), we find no
significant variations in performance data and conclude that RAID level has no impact on user I/O
workloads placed on the Exchange Server in this test environment.
Table 3. User IOPS in RAID 5 drive configurations
Processor utilization
Table 4 shows processor utilization and workload for each storage device and drive in RAID 10
configurations. This comparison determines if a particular storage device or drive configuration has
any significant impact on processor utilization. The performance data is derived by capturing data
from the middle 4 hours of an 8-hour performance test. The values show both the average and the
95th percentile of each storage device and drive configuration. The percentage of processor time
13
reveals an average of 7.33% utilization, with average highs estimated at 16.33%. The standard
deviation between each test configuration was +/--- .471%.
We recommend a four-core processor configuration for a multiple-role (Hub, CAS, and Mailbox
roles) Exchange 2007 Server. To accommodate CAS and Hub utilization on a single server with the
Mailbox role, reduce the 1,000 mailbox per core calculation based on the average client profile by
20% (800 mailboxes/core).
95th percentile 18 18 15 15 14 18
Table 5 shows the results of processor utilization based on RAID 5 storage configurations. The
percentage of processor time shows an average of 7.16% processor utilization, with the average high
estimated at 17%. The standard deviation between each test configuration was +/--- .372%.
This data shows a consistent consumption of processor utilization for all storage test configurations.
The data indicates that a specific storage device and a properly sized drive configuration have no
impact on the processor utilization of the Exchange Server. The tests also confirmed that there were
no significant variations in processor utilization between RAID levels; the RAID level on a properly
sized configuration has no impact on the processor utilization of the Exchange Server.
95th percentile 18 19 15 18 15 17
Memory utilization
Table 6 shows the memory utilization and workload for each test configuration in RAID 10
configurations. Microsoft recommends 8 GB of memory for a multiple-role Exchange 2007 Server,
and an additional 2 MB to 5 MB per mailbox. Based on this recommendation, 13 GB (8 GB per
14
server plus 5 GB per mailbox x 1,000) was planned for the test environment. To accommodate the
memory requirement, the Exchange Server required 16 GB of RAM.
Table 6 identifies any significant fluctuation in memory utilization on the Exchange Server resulting
from changes in storage devices and drive configurations. The data is derived by evaluating the
middle 4 hours of an 8-hour performance test.
The Exchange database cache indicates an average size of 4,040 MB with fluctuations as high as
5,164 MB, and a deviation of +/– 151 MB per performance test. The amount of memory committed
by the Exchange Server reflects an average of 7,005 MB with an average high of 7,927 MB and a
standard deviation of +/– 182 MB between test configurations.
The page file information shows consistent paging between all test configurations and indicates an
average of 69.07 MB (+/– 3.67) of Pool Nonpaged Bytes used by the operating system and kernel
processes. The amount of Pool Paged Bytes used for applications was an average of 79.17 MB, with
a deviation of +/– 2.98. This memory utilization data indicates that selecting a properly-sized
storage device and drive configuration does not have any significant impact on memory utilization on
the Exchange 2007 Server.
Database cache size avg 4,187 3,961 4,134 3,915 4,230 3,816
Page/sec avg 27 31 26 39 58 26
95th percentile 67 70 64 57 67 59
Note
Pool Paged Bytes and Pool Nonpaged Byte counts were converted to megabytes for simplicity.
15
Table 7 shows the results of memory utilization based on RAID 5 storage configurations. The
Exchange database cache size shows an average size of 3,836 MB with fluctuations as high as
5,044 MB, and a deviation of +/– 143.14 MB per performance test. The amount of memory
committed by the Exchange Server shows an average of 6,830 MB, with an average high of 7,919
MB and a standard deviation of +/– 213 MB between test configurations.
The page file information reveals consistent paging between all test configurations, and an average
of 69.96 MB (+/– 4.19) of Pool Nonpaged Bytes used by the operating system and kernel processes.
The amount of Pool Paged Bytes used for applications and non-system processes was an average of
81.94 MB with a deviation of +/– 10.45. There were no significant changes in memory utilization for
any storage device and drive configuration. The evaluation and comparison of RAID types shows that
RAID level has no impact on memory utilization.
Table 7. Memory utilization in RAID 5 drive configurations
Database cache size avg 3,677 3,795 3,913 3,896 4,073 3,662
Page/sec avg 25 52 28 46 28 39
95th percentile 62 71 74 63 73 61
Note
Pool Paged Bytes and Pool Nonpaged Byte counts were converted to megabytes for simplification.
Prior to putting the Exchange Server into simulated production (LoadGen testing), we evaluated the
Exchange 2007 physical disks using the Microsoft Exchange Jetstress utility to verify the performance
and stability of the disk subsystem. Jetstress verifies if an Exchange Server disk configuration meets
or exceeds a simulated user I/O load criterion in a nonproduction environment. The Jetstress utility
identifies the maximum I/O on a given storage configuration. For this test (1,000 users with an
16
estimated heavy user profile of .31 IOPS per user), the Jetstress target criterion were 310 transfers per
second (number of users X user profile) on the database disks:
Table 8 and Table 9 provide the Jetstress physical disk results for each storage device and disk
configuration in both RAID 10 and RAID 5 testing. This performance data is derived from the
evaluation of an 8-hour Jetstress performance test.
Table 8 shows Jetstress results of each storage device and drive configuration based on RAID 10
configurations. This test data shows the overall performance capabilities of the SAS, SATA, and SCSI
technologies, and their ability to meet or exceed the target criterion of 310 IOPS for database disks.
The SATA configuration shows that the SATA drive configurations produced the least amount of IOPS
in (transfers per second) and maintain the recommended Exchange read and write latencies (Disk
Sec/Read Disk Sec/Write), which is less than 20 ms for database disk and less than 10 ms for Log
File Disks. In addition, the MSA1500 with 8 X 500 GB 7.2 K (MSA20 enclosure) configuration falls
short of test criterion of 310 IOPS on the database disk. This configuration could result in a physical
disk bottleneck if deployed in a 1,000-user production environment. The MSA20 is a SATA enclosure
with Ultra 320 SCSI host connectivity, ideal for low-cost, high-capacity minimum I/O requirements,
and is not intended for high transactional applications such as Exchange.
Comparisons of the MSA60 and MSA70 SATA configurations show that both configurations meet
the test criterion. However, the MSA70 with 30 X 120-GB 5.4 SATA drives produced a significantly
higher number of IOPS (1,414) than the MSA60 with 8 X 500-GB 7.2K SATA drives (352). This can
be attributed to the increased number of spindles in the MSA70 SATA configuration.
The SAS and SCSI drive configurations show a much higher number of I/O operations and
lower read and write latencies. This data shows clearly that SAS and SCSI drives will produce
higher I/O capabilities than SATA drives. SATA drives are more affordable in comparison to
SAS and SCSI hard drives and may be a more attractive solution to some administrators with
limited budgets common to small to mid-range business environments. Administrators with limited
budgets who are examining SATA drive solutions might consider using a larger number of smaller
capacity drives, rather than fewer high capacity drives to achieve better I/O capabilities. See
“Hard drive technologies” on page 48.
17
Table 8. Jetstress performance in RAID 10 drive configurations
Logfile Disk/sec write avg .008 .000 .009 .001 .008 .003
Table 9 shows the Jetstress results of each storage device and drive configuration based on Raid
5 configurations. The SAS and SCSI configurations performed well and achieved or exceeded the
Jetstress I/O test criterion of 310 IOPS.
The SATA test data shows that the SATA drive configurations produced the fewest IOPS while
maintaining the recommended Exchange read and write latencies. As in Table 8, the MSA1500
with 11 X 500-GB 7.2 K (MSA20 enclosure) hard drive configuration fell short of the test criterion
of 310 IOPS on the database disk. This configuration could result in a physical disk bottleneck if
The 500-GB SATA drives in the MSA60 and MSA1500 configurations required three additional hard
drives, based on the results from the HP Storage Planning Calculator. The SATA drive performance
limitations and compensation for the RAID 5 write penalty requires additional drives. It would be
more economical (and provide higher redundancy) to use the RAID 10 SATA configurations.
Both the MSA60 and MSA70 SATA configurations meet the test criterion. The SAS and SCSI drive
configurations show a much higher number of I/O operations, and lower read and write latencies.
This data shows that SAS and SCSI drives produce higher I/O capabilities than SATA drives.
deviations for performance. The P800 Smart Array Controller performs the write caching and
compensates for the RAID 5 overhead. See “Hard drive technologies” on page 48.
18
Table 9. Jetstress performance in RAID 5 drive configurations
95th percentile
95th percentile
Logfile Disk sec/write avg .001 .000 .000 .000 .001 .003
Table 10 and Table 11 provide the physical disk performance data from the simulated production
environment using the Microsoft Exchange Load Generator. This data shows storage device
performance under the simulated Exchange load. The data is derived from the middle 4 hours of
an 8-hour performance test.
Table 10 shows the results of the Exchange database physical disk performance based on RAID 10
storage configurations. The Exchange database physical disk evaluation shows that the Exchange
Server sustained a consistent workload across all storage devices and drive configurations, averaging
122.66 transfers per second with an average high of 285.83 transfers per second, and a deviation
of +/– 4.21 transfers per second between each performances test. The Disks read and write latencies
(Disk Sec/Read and Disk Sec/Write) are within Microsoft Exchange 2007 recommendations with
the exception of the MSA1500 SATA configuration. Microsoft recommends an average of less than
20 ms for read latencies, and less than 10 ms for write latencies.
The SATA performance data evaluation shows that the SATA configurations maintained the same
amount of transfers per second on the database disk and stayed within Exchange performance
guidelines. However, by comparison the read and write latencies and average disk queue length
19
(number of operations pending) of the SATA configuration is higher than the SAS and SCSI
configurations. If an administrator chooses to implement a properly sized and provisioned SATA
Exchange solution, the impact will be higher latencies and higher read and write queue lengths.
95th percentile 35 28 25 27 38 30
Disk read MB/s avg .28 .44 .30 .44 .34 .34
Ave Disk sec/read avg .014 .006 .014 .007 .024 .006
Disk read queue avg .244 .121 .334 .156 .477 .105
Disk write MB/s avg 1.70 1.74 1.70 1.79 1.93 2.17
Note
Disk Bytes Read and Write performance counts were converted to megabytes for simplicity.
Table 11 shows the results of the Exchange log file physical disk performance based on RAID 10
storage configurations. The Exchange log file physical disk evaluation shows that the load on the
Exchange Server sustained a consistent workload across all storage devices and drive configurations,
averaging 75.66 transfers per second with the average high of 161.83 transfers per second, and a
deviation of +/– 2.42 transfers per second between each performance test. Each storage group and
drive configuration was within Microsoft Exchange recommended performance guidelines for write
latencies (10 ms). No read operations are conducted on log file physical disks with Exchange 2007
(unless LCR/CCR is implemented).
20
Log file operations consist of sequential writes that commit data from cache to the logs. By separating
databases and log files and placing them on dedicated physical disks, you can improve performance
by splitting random and sequential I/O operations. Because log file generation is a write-intensive
operation, place the log files on a RAID 10 volume if possible. However, it is acceptable to place
databases on RAID 5 volumes if the storage operating budget is restricted.
Table 11. Physical disk log file performance in RAID 10 drive configurations
Transfers/sec avg 71 75 76 79 76 77
95th percentile
95th percentile
95th percentile
95th percentile
Disk write MB/s avg .53 .47 .53 .53 .51 .53
Disk write queue length .063 .011 .041 .018 .068 .067
Disk queue length avg .063 .011 .041 .019 .068 .067
Note
Disk Bytes Read and Write performance counts was converted to megabytes for simplicity.
Table 12 and Table 13 show the Exchange database physical disk performance in the simulated
production environment.
Table 12 shows the Exchange database physical disk performance based on RAID 5 storage
configurations. The resulting evaluation shows that the load on the Exchange Server sustained a
consistent workload across all storage devices and drive configurations, averaging 125.33 transfers
per second with the average high of 291.5 transfers per second, and a deviation of +/– 6.39
21
transfers per second between performances test. The disks read and write latencies (Disk Sec/Read
and Disk Sec/Write) were within the Microsoft Exchange 2007 recommendations, with the exception
of the MSA1500 SATA configuration, consistent with the RAID 10 database tests results.
As in the RAID 10 database disk results, a comparison of the read and write latencies and average
disk queue length (number of operations pending) reveal that the SATA configurations are higher
than the SAS and SCSI configurations. The RAID 10 and RAID 5 database physical disk performance
data reflect no significant performance impact between the storage devices and disk types. The RAID
5 write overhead was compensated by the write cache of the P800 Smart Array Controller.
95th percentile 57 54 43 45 43 38
Disk read MB/s avg .39 .33 .33 .32 .42 .37
Disk read queue length .342 .124 .345 .122 .555 .130
Disk write MB/s avg 1.76 1.75 1.78 1.71 1.83 2.01
Disk write queue length .093 .074 .179 .058 .079 .929 .540
Note
Disk Bytes Read and Write performance counts were convert to megabytes for simplicity.
Table 13 shows the Exchange log file physical disk performance data based on RAID 5 storage
configurations. The evaluation shows that the Exchange Server sustained a consistent workload
across all storage devices and drive configurations, averaging 79.83 transfers per second with an
average high of 168.66 transfers per second and a deviation of +/– 4.05 transfers per second
22
between each performances test. Each storage group and drive configuration was within the
Microsoft Exchange recommended performance guidelines for write latencies (10 ms). There were no
significant performance deviations between RAID 10 and RAID 5 log file physical disks.
Transfers/sec avg 76 82 77 83 86 75
95th percentile
95th percentile
95th percentile
95th percentile
Disk write MB/s avg .53 .58 .50 .51 .71 .51
Disk write queue length .028 .017 .016 .18 .023 .072
Disk queue length avg .028 .017 .016 .18 .023 .022
Exchange performance
Table 14 and Table 15 provide the actual Exchange performance characteristics during each test
configuration in both RAID 10 and RAID 5 configurations. This performance data was derived from
the middle 4 hours of an 8-hour performance test.
Table 14 shows the number of messages sent and received (delivered), and the number of messages
held in queue during testing, based on RAID 10 configurations. In addition, the remote procedure
call (RPC) operations determined the end-user experience, and identified any performance latencies
that could occur when accessing the Exchange Server using Microsoft Outlook.
The number of delivered messages reveals a consistent workload placed on the Exchange 2007
Server by the Exchange Load Generator clients. It shows an average of 52,329 messages received,
with a deviation of +/– 2,228 messages (4%) at a rate of 3 messages per second between test
23
configurations. The number of sent messages shows an average of 14,374, with a deviation of +/–
1,606 messages (11%) at a rate of .80 messages per second.
An assessment of RPC operations shows a consistent user environment across all test configurations.
The average number of RPC operations from all six test configurations was 390 operations with a
standard deviation of +/– 40.51 (10.38%) between each storage device and drive configuration.
The RPC averaged latency in milliseconds and is based on the average for the last 1,024 packets.
Microsoft Exchange performance guidelines recommend that this value stay below 50 ms at all times.
The RAID 10 performance data evaluation shows performance well below the 50-ms threshold for all
storage device configurations. However, closer evaluation of the SATA disk configurations reflects
higher number of messages queued for submission and higher RPC latencies when compared to the
SAS and SCSI drive configurations. The selection of SATA drives in an Exchange 2007 configuration
led to higher latencies and increased message queue lengths compared to SAS and SCSI drive
configurations.
95th percentile 13 12 12 12 11 7
95th percentile 2 2 2 2 2 2
Msgs opened/sec 13 13 14 13 13 13
95th percentile 35 33 35 33 31 26
95th percentile 0 0 0 0 0 0
95th percentile 2 2 2 2 2 2
95th percentile 30 6 17 4 30 6
24
Note
Pool Paged Bytes and Pool Nonpaged Bytes counts were converted to megabytes for simplicity.
Table 15 shows the number of messages sent, received, or held in queue during each test iteration,
based on RAID 5 configurations. The number of messages delivered (received) reveals a consistent
workload placed on the Exchange 2007 Server by the Exchange Load Generator clients. The data
shows an average of 45,937 received messages, with a deviation of +/– 2,132 messages (4.64%) at
a rate of 3 messages per second between test configurations.
An average of 12,312 messages was sent, with a deviation of +/– 513 messages (4.16%) at a rate of
.82 messages per second. The assessment of RPC operations shows a consistent user environment
across all test configurations. The average number of RPC operations from all storage configurations
was 415 operations, with a standard deviation of +/– 14.73 (3.45%) between each storage device
and drive configuration. The RPC averaged latency evaluation shows performance well below the
50-ms threshold for all storage device configurations.
As in the RAID 10 Exchange results, the SATA disk configurations indicate more messages queued
for submission and higher RPC latencies when compared to the SAS and SCSI drive configurations.
This concludes that selecting SATA drives in an Exchange 2007 configuration would lead to higher
latencies and increased message queue lengths compared to SAS and SCSI drive configurations.
In addition, the comparisons of Exchange performance data in RAID 10 (Table 14) and RAID 5
(Table 15) reflect similar performance, and indicate no significant change in Exchange performance
based on RAID level.
25
Table 15. Exchange performance in RAID 5 drive configurations
Msgs delivered/sec 3 3 3 3 3 3
95th percentile 13 13 12 13 12 12
95th percentile 2 2 2 2 2 2
Msgs opened/sec 13 13 13 13 13 13
95th percentile 35 34 32 31 32 31
95th percentile
95th percentile 2 2 2 2 2 2
95th percentile 18 5 16 4 26 9
Testing summary
The tests provide mid-range customers (1,000 Exchange users) with insight into storage planning for
Exchange 2007. They include the technical data needed to help you make informed decisions
regarding drive technologies and RAID levels that suit business needs and operational budgets.
Because of the lower cost associated with SATA-based storage, many administrators find SATA
technologies a more feasible alternative to the more expensive SAS and SCSI options, especially
in companies with smaller environments and operating budgets.
After exploring the impact of using SATA drive solutions in an Exchange 2007 environment, it is
evident that a properly sized SATA configuration can satisfy Exchange storage requirements for most
small to mid-range businesses. In some cases, such as SATA configurations using the MSA1500 and
MSA20 enclosure, it does not meet Exchange 2007 performance guidelines, and therefore is not an
HP recommended solution.
26
Proper sizing of the Exchange storage solution is important, and planning should be focused
on performance over capacity. The HP Storage Planning Calculator is a useful tool for storage
planning and provides results based on disk controller, storage device, and drive type. In this
test environment, the HP SPC provided storage results that performed well and met the Microsoft
Exchange performance guidelines. Testing showed that the SATA configurations could meet or
exceed Exchange performance guidelines at the cost of less I/O capability, higher read and write
latencies, and longer message queues. Testing MSA1500 with MSA20 SATA drives is not listed as a
valid HP storage configuration in the HP Storage Planning Calculator. A drive configuration in one
storage device may not meet performance requirements in another storage device.
When selecting a storage solution, it is important to evaluate the dependability and anticipated life
span, described as the mean time between failures (specifically the mean time between failures
(MTBF) of a system expressed in hours based on 24 x 7 operations). While SATA drives can save
cost for some Exchange deployments compared to SAS and SCSI, the MTBF is considerably less.
MTBF for a SATA drive is estimated at 600,000 hours, and 1.4 million hours for SAS drives (results
may vary based on duty cycle). Organizations that consider their Exchange capabilities as a vital
business function, or have stringent service license agreement (SLA) system recovery requirements,
may not view SATA as a viable option. Whichever drive technology best suits your Exchange
environment, success depends on sizing and testing the solution before deployment.
27
Best practices and results
This section describes best practices for IT professionals who operate and support small to mid-range
business environments. It describes the major differences between Exchange 2007 and previous
versions, and provides information about infrastructure planning, deployment, and migration to
Exchange Server 2007.
This section also shows the results of testing the MSA products and drive performance to assist in
determining a best storage solution for the small to mid-range business environment. Individuals not
familiar with Exchange 2007 should refer to Microsoft Exchange Server 2007 for an introduction to
new features and changes in the Exchange 2007 architectural design.
This section also includes recommendations for customers who plan to upgrade from previous
versions of Exchange, or migrate from other messaging systems. The following list of best practices
and recommendations should be considered by Exchange and system administrators to successfully
deploy Exchange in their messaging environments.
Exchange administration
• Planning: Before you deploy Exchange 2007, plan properly to ensure smooth adaptation into
the organizational environment. Begin by assessing the business requirements and the current
IT computing environment. Analyze the Active Directory logical and physical topologies to
determine which Exchange 2007 organizational topology best fits your environment size and
messaging needs. In addition, determine whether to adapt a centralized or a decentralized
administrative model.
• Exchange Jetstress: After determining the Exchange 2007 physical disk I/O requirements, use the
Jetstress tool to verify that the design disk performance can meet the expectations of the anticipated
production environment. Do this in a test lab prior to the production environment deployment.
• Storage planning tools: Available storage planning tools include an Exchange Storage Planning
Calculator, an HP Storage Planning Calculator (SPC), and a Sizing and Configuration Tool.
You can use an Exchange Storage Planning Calculator to simplify planning. Microsoft provides
an Exchange 2007 Mailbox Server Role Storage Requirements Calculator spreadsheet that
can be downloaded from:
http://msexchangeteam.com/files/12/attachments/entry438481.aspx
HP has developed the HP Storage Planning Calculator (SPC) for Microsoft Exchange Server 2007.
This tool was designed to provide guidelines for disk, storage controller, and enclosure planning
for successful Exchange Server 2007 deployments; it is available for download from:
http://h71019.www7.hp.com/activeanswers/Secure/511755-0-0-0-121.html
HP also provides the HP Sizing and Configuration Tool for Microsoft Exchange Server 2007. This
tool provides HP ProLiant server and storage sizing guidelines for Microsoft Exchange Server
2007 solutions and can be downloaded from:
http://h71019.www7.hp.com/activeanswers/Secure/483374-0-0-0-121.html
28
Performance and recovery
To optimize performance and provide fault tolerance, place databases and logs on separate
physical disks. This enables recovery of data to the point of failure. If the log files and databases
are maintained on the same disk, and the disk fails, recovery to the point of the last backup is the
only possibility.
Performance is also improved because database I/O operations are random, whereas log file
operations are sequential. The combination of both on the same physical disk degrades I/O
performance. The following files should be partitioned on separate disks:
Storage administration
Drive selection for an Exchange 2007 deployment depends on various environmental factors, and
requires careful considerations, such as budget, business requirements, and I/O requirements. SATA
drives may seem like an affordable solution; if properly sized, they can support an organization’s
Exchange 2007 storage requirements. However, other factors, such as MTBF and the shorter life
cycles of SATA drives, should be considered.
Organizations that consider Exchange Servers as a mission-critical resource should examine the
increased durability of SAS and SCSI drives. The evolutionary replacement of SCSI with SAS is strong
in many environments. This is driven by the SAS technology’s advantages, such as quadrupling SCSI
bandwidth for substantial performance gains, and accommodating up to 128 SAS devices on a single
controller, as compared with 16 for parallel SCSI. In addition, the ability to mix high-performance
SAS and low-cost SATA drives on the same backplane provides unprecedented flexibility.
The trade-off of the lower cost for the SATA drives can also mean lower I/O capabilities. Even though
a SATA configuration may meet the organization’s Exchange 2007 user I/O requirement, consider
how the reduced I/O will impact backup and recovery and the organization’s SLAs.
Improper storage design leads to bottlenecks and poor performance. It is important to meet storage
capacity requirements for an Exchange deployment. To ensure proper performance, the proposed
solution must have the correct number of spindles to meet the I/O requirements. Use the following
formula to determine the total number of spindles for RAID 10:
1.25 × [(Mailboxes × IOPS per mailbox/IOPS per spindle) + %Read I/O]/[%Reads I/O +
(%Write I/O/2)] = Spindles
Server administration
The Microsoft Exchange Server Best Practices Analyzer (ExBPA) Management Pack works with
Exchange Server 2003 to monitor performance after server deployment. The Management Pack
29
includes a tool that runs on the server to identify configuration issues which could result in poor
performance, scalability and unplanned downtime. For more information about the ExBPA tool, see
http://technet.microsoft.com/en-us/exchange/bb288481.aspx
For more information about server requirements, see Microsoft Exchange Server 2007 .
30
Conclusion
For customers who are migrating or planning deployment to Exchange Server 2007, this paper
provides technical guidance, best practices, and supporting data for businesses that support 1,000
users on a single array.
Experienced HP Exchange engineers have thoroughly tested several arrays and disk types in the
MSA array family. The insights reported here can help you evaluate the performance data and
implement the proven best practices, ensuring a trouble free environment. Among the best practices
identified during testing are:
• Tools are critical. Use sizing tools to properly plan the environment, and use the Jetstress tool to
verify that the proposed design will meet performance expectations.
• Separate databases and logs. Place databases and logs on separate physical disks to optimize
performance and provide fault tolerance.
• HP recommends using RAID 10. For the best performance, use RAID 10 for both Exchange
databases and log files. If you have budget constraints, consider RAID 5 for the database disks,
but always place the Exchange log files on a RAID 10 volume.
• SATA is an affordable option. The trade-off is a lower life cycle, less reliability, and a potential
impact to SLAs.
• SAS/SCSI is the choice for mission-critical messaging environments. With its higher I/O
capabilities, SAS/SCSI offers the most reliability.
This white paper has identified the proven technical best practices. Evaluating the presented trade-offs
can help you implement a solution that satisfies your budget and organizational priorities.
In order to develop technical materials that address your information needs, we need your feedback.
We appreciate your time and value your opinion. The following link will take you to a short survey
regarding the quality of this paper:
http://hpwebgen.com/Questions.aspx?id=12046&pass=41514
31
Appendix A. Bill of materials
Storage Devices
32
Appendix B. HP Storage Planning Calculator
The tests described in this document are presented to help you evaluate the performance
characteristics of SAS, SATA, and SCSI drive technologies and identify the impacts of an Exchange
2007 deployment in a direct storage configuration. The HP Storage Planning Calculator (SPC) for
Exchange 2007 was designed to determine the best configuration for each storage device and drive
type needed to support an Exchange user workload. This tool ensured the proper sizing for each
drive technology to accommodate our Exchange test environment.
The testing also provided an opportunity to validate the accuracy of the SPC by comparing the
recommended solutions with the actual performance data. Tests indicated that all the storage
recommendations from the SPC met the Exchange 2007 storage requirements within Exchange
performance guidelines and with consistent results.
The MSA1500 tests (using the MSA20 with 500-GB SATA) did not meet performance testing criteria.
However, that configuration solution was not listed in the HP Storage Planning Calculator; it was
added to the tests to validate HP’s recommendation that the MSA20 be used only for file storage,
disk-to-disk backup, and data archiving. These tests also indicate that the same drive configuration
can provide satisfactory results for one storage device (MSA60 with 500-GB SATA), but may perform
poorly for another storage device (MSA1500 with 500-GB SATA).
Figure 6, Figure 7, Figure 8, Figure 9, and Figure 10 show the HP Storage Planning Calculator’s
recommended configurations based on the Exchange user workloads.
33
Figure 6. MSA60 with 300-GB 15K RPM SAS 3.5
34
Figure 7. MSA60 with 500-GB 7.2K RPM SAS 3.5
35
36
37
38
As you consider whether to move a current Microsoft Exchange Server 2003 implementation to
Microsoft Exchange Server 2007, it is important to understand the application’s new capabilities and
their implications for your current infrastructure.
With Exchange 2003, the limitations on available memory created a performance bottleneck.
Exchange 2003 uses a maximum of 3 GB of virtual address space (using the 3-GB switch); this limits
the number of users on a mailbox server and the size of the mailboxes.
Exchange Server 2007 uses a 64-bit environment, allowing the server to take advantage of the
database cache and additional server RAM. With previous Exchange versions, memory limitations
increased the probability that the application would read data from disk, thus increasing disk I/O
and impacting performance. The larger memory capabilities in Exchange Server 2007 minimize
database I/O and improve the overall performance.
Role-based servers
Microsoft Exchange Server 2007 introduces role-based deployment. The assigned server roles allow
the required features and components to perform a specific function in the messaging environment.
Depending on how a messaging system will be deployed and distributed, distinct server roles can be
configured as follows:
• Client Access Server (CAS). The CAS role is responsible for all client interaction, with the
exception of the direct Messaging Application Programming Interface (MAPI) remote procedure
call (RPC). It includes interactions with Outlook Web Access, Outlook Anywhere, POP, IMAP,
ActiveSync, and Web Services.
• Hub Transport (Hub). The Hub role is responsible for all routing; it provides better journaling for
improved compliance and improved availability.
• Mailbox Server. The Mailbox Server role is the MAPI RPC end point. It stores all mailbox items,
searches for performance improvements, and provides the high-availability features of Local
Continuous Replication (LCR) and Cluster Continuous Replication (CCR). The Mailbox Server role
makes use of the recommended maximum of 32-GB memory and supports up to 50 storage
groups (SGs) with an improved checkpoint depth.
• Unified Messaging (UM). The UM role enables remote access services such as Fax, voice, and
cell access; it is not a messaging system requirement.
39
• Edge Transport. The Edge Transport Server role was designed to handle Internet-facing SMTP
relay and provides increased security; it is not a messaging system requirement. The Edge
Transport role requires the use of a dedicated server that is not in the Active Domain.
The CAS server role, Hub server role, and Mailbox server role can coexist on a single server in any
combination. The assignment of roles is based on capacity and performance requirements.
Because it requires an investment in new 64-bit hardware, the move to Microsoft Exchange Server
2007 presents an opportunity to consider the potential impact of the reduced I/O profiles on the
total infrastructure.
• With more data and instructions available for processing in RAM, fewer servers may be needed
because each server can host more mailboxes. Of course, sufficient memory needs to be allotted
for other CPU processes, such as antivirus and recovery.
• It may be possible to deploy lower performing storage solutions because of the lower burden
placed on disk I/O performance. In some cases, DAS may provide sufficient performance. In
others, organizations may find that the benefits of centralized storage management using a
storage area network (SAN) outweigh the cost savings of a DAS solution.
• With the constantly increasing dependence on messaging systems and the use of larger
attachments, organizations can expect that mailbox databases will continue to grow. Although
Exchange 2007 supports larger mailbox sizes, there is still a need to balance user demands
against infrastructure considerations such as database manageability and server workload.
Microsoft is encouraging Exchange 2007 administrators to use its embedded, host-based Volume
Shadowcopy Service (VSS) technology to perform backups. However, many users will choose to
continue to deploy online streaming backups, either as a first step in implementation or as a valid
method for simplifying processes or leveraging concurrency.
VSS technology is critical for applications that require consistent, point-in-time copies of open files
and databases. With Exchange 2007, VSS backups can be restored to an Exchange Recovery
Storage Group on any Exchange Server in the organization.
Continuous replication
With its built-in, host-based continuous replication capabilities, Exchange 2007 provides log shipping
services to remote storage or remote hosts, which enhances overall availability. Local Continuous
Replication (LCR) allows the database to be replicated to alternate storage on the same host, with log
replay keeping the passive database up-to-date. Cluster Continuous Replication (CCR) provides for
40
logs to be shipped to a passive cluster node where the replicated logs are replayed into the passive
database. In addition to quick recovery, CCR provides important benefits in other backup and
recovery operations:
• Ability to perform daily incremental backups with weekly full backups from a remote data copy
• Ability to minimize the reliance of backup windows by backing up the passive LCR or CCR
copy of the database
• Ability to perform off-host backups with the passive cluster node in CCR
Microsoft has made many changes to Exchange Server to address customers’ ongoing needs
for functionality and performance. HP has done extensive testing to determine how these new
capabilities impact the storage configuration and has defined best practices for implementing
Microsoft Exchange Server 2007 with HP server and storage products.
41
Appendix D. Microsoft Exchange Load Generator
The Microsoft Exchange Load Generator (LoadGen) is a simulation tool that measures the impact of
MAPI clients on the Exchange Server. It measures the Exchange Server response to a configured
user workload.
In our testing, we used LoadGen to simulate the workload of 1,000 users with 750-MB mailboxes on
the Exchange Server in each storage configuration. We developed a customized configuration that
would populate mailboxes to the appropriate size and produce the correct amount of messaging
activity to simulate a heavy user profile. User activities included send and receive messages, browse
the calendar, browse public folders, and download offline address books. For our test, the modified
LoadGenConfig.xml yielded an average mailbox size of 750 MB per user.
Figure 11 shows the user profile action in each test simulation and identifies the type and number of
mail activities conducted for each user.
42
Figure 11. Action profile
Figure 12 shows the successful completion of an 8-hour LoadGen test run that performed 130,492
tasks with no task exceptions. The results show 132 tasks per user day for 1,000 users.
Figure 13 and Figure 14 show the amount of mail that was sent and received (24,000 messages
sent and 87,000 messages received). The averages were 24 sent and 87 received per user; these
averages fall within the test criteria for a heavy user profile (estimated at 20 sent and 80 received).
43
Figure 12. Load Generator test results
44
45
Appendix E. Software
• Windows Server 2003 R2 Enterprise X64 Edition (Build 3790: Service Pack 2) (Exchange Servers)
• Windows Server 2003 Enterprise Edition (Build 3790: Service Pack 2) (Domain
Controllers/LoadGen Clients)
Software updates
• Hotfix for Windows Server 2003 (KB926139)
46
• Update for Windows Server 2003 (KB932596)
47
Appendix F. Hard drive technologies
SATA or Serial Advanced Technology Attachment (ATA) is the predecessor of the traditional Parallel
Advanced Technology Attachment (PATA) that is common in most workstations and personal
computers. PATA, also called Integrated Drive Electronics (IDE) was the standard bus interface on
the original IBM AT computer. ATA is the official American National Standards Institute (ANSI)
standard term.
Most motherboards include two ATA 40-pin connectors; each is capable of supporting two devices
(one master and one slave). SATA is based on serial signaling technology, unlike current IDE hard
drives that use parallel signaling. SATA offers performance as high as 3.0 Gb/s per device. Because
SATA uses only four signal lines, it allows for much more compact (and less expensive) cables as
compared with PATA. It also offers features such as hot-swapping and Native Command Queuing
(NCQ).
The SATA port multiplier contains provisions for many drives to be connected to the same SATA, unlike
the master/slave limitation of PATA. In addition, SATA drives may be plugged into Serial Attached
SCSI (SAS) controllers and communicate on the same physical cable as native SAS disks. SAS disks,
however, may not be plugged into a SATA controller.
SCSI is a high-performance peripheral interface that can independently distribute data among
peripherals attached to the computer. Unlike ATA, SCSI incorporates the instructions needed to
communicate with the host computer. As a result, the host computer is more efficient in performing its
user-oriented activities.
SCSI is a specification for a peripheral bus and command set defined in ANSI standard
X3.131-1986. SCSI drives are usually more suitable for high-end computer systems that require the
maximum possible performance. SCSI provides for higher data transfer rates and less CPU load than
ATA but has higher cost and complexity in the setup. Also, SCSI supports more devices than ATA.
Another important advantage of SCSI is that most SCSI products are backward-compatible. A faster,
newer drive can still work with the older and slower controller but with less performance.
SAS evolved as the new serial-based technology when the SCSI hard drive reached its physical speed
threshold of 320 MB/s. SAS technology extends the benefits of parallel SCSI, such as high reliability,
enterprise-class performance, and manageability; it also provides an increase in I/O performance.
The SAS standard uses a hard-drive interface that is seamlessly compatible with both existing SATA
and SAS hard drives, thereby allowing a user to mix SAS and SATA drives in the same system for
greater flexibility in deploying storage solutions. Another capability of SAS allows users to attach
up to 128 SAS storage devices, an improvement from the 16 devices that could exist in a SCSI
environment.
48
For more information
This section lists references and their online locations.
Note
Some of the following links are secure websites that require an HP Passport registration. HP Passport is a single login service
that lets you register with HP Passport-enabled websites using a single user identifier and password of your choice.
http://h71028.www7.hp.com/enterprise/cache/324315-0-0-0-121.html
http://h71019.www7.hp.com/activeanswers/Secure/511755-0-0-0-121.html
http://h71019.www7.hp.com/activeanswers/Secure/483374-0-0-0-121.html
http://h18004.www1.hp.com/storage/disk_storage/msa_diskarrays/san_arrays/msa1500cs/
index.html
http://h18002.www1.hp.com/storage/disk_storage/msa_diskarrays/drive_enclosures/msa60/
qa.html
http://h18002.www1.hp.com/storage/disk_storage/msa_diskarrays/drive_enclosures/msa70/
index.html
http://h10010.www1.hp.com/wwpc/us/en/sm/WF05a/15351-15351-3328412-241644
241475-1121516.html
http://h10010.www1.hp.com/wwpc/us/en/sm/WF05a/15351-15351-3328412-241644
241475-3201178.html
http://technet.microsoft.com/en-us/library/aa996058.aspx
http://msexchangeteam.com/archive/2007/01/16/432222.aspx
49
• Exchange 2007 Mailbox Server role storage requirements calculator spreadsheet
http://mseschangeteam.com/files/12/attachments/entry438481.aspx
http://www.microsoft.com/downloads/details.aspx?familyid=ddec1642-f6e3-4d66-a82f
8d3062c6fa98&displaylang=en
http://technet.microsoft.com/en-us/library/bb738142(EXCHG.80).aspx
50