Enhancing performance of HP ProLiant servers containing
Smart Array 6i controllers
best practice
Abstract.............................................................................................................................................. 2
Introduction......................................................................................................................................... 2
Factors affecting array controller performance......................................................................................... 2
RAID stripe size ............................................................................................................................... 2
Benchmarks used ............................................................................................................................. 3
Bus speed ....................................................................................................................................... 4
Recommendations................................................................................................................................ 4
For more information............................................................................................................................ 6
Call to action ...................................................................................................................................... 6
Abstract
Benchmark testing in HP laboratories has shown that with optimized configuration settings, HP
ProLiant servers with Smart Array 6i controllers can match or exceed the performance of ProLiant G3
servers using Smart Array 5i controllers. This paper presents results of that testing and defines some
best practices for upgrading from Smart Array 5i controllers and for optimizing performance of
ProLiant servers containing Smart Array 6i controllers.
Introduction
The HP Smart Array 6i controller is an Ultra320 intelligent array controller for entry-level, hardwarebased fault tolerance. At this writing, it is integrated on the system board of ProLiant DL360 G4/G4p,
ProLiant DL380 G4, ProLiant DL385, ProLiant DL580 G3, and ProLiant BL20p G3 servers. HP
designed this controller specifically to provide a highly cost-effective alternative to software-based
RAID for protection of OS, applications, and logs.
In response to inquiries about performance of Smart Array 6i controllers, HP launched a program to
compare performance of Smart Array 6i controllers with performance of Smart Array 5i controllers
and to analyze possible affects on storage system performance in various computing environments.
This paper summarizes results of benchmark tests conducted in HP laboratories. It also defines some
best practices for upgrading from Smart Array 5i controllers and for optimizing performance of
ProLiant servers containing Smart Array 6i controllers.
Factors affecting array controller performance
HP engineers tested several benchmarks on different server configurations using Smart Array 5i and
Smart Array 6i controllers. The test configurations were selected from customer cases submitted for
review. In many cases, a simple change of either the RAID stripe size or the test benchmark improved
the performance of the Smart Array 6i controller. In other cases, due to the nature of the application
workloads, there was little or no performance difference between the Smart Array 5i controllers and
the Smart Array 6i controllers.
The tables in this section summarize results obtained in four types of tests conducted in HP
laboratories. Each test was run five times. The tables show the average results from the respective test
series.
RAID stripe size
HP tests revealed that one significant factor affecting performance of the Smart Array 6i controller is
the drive data block size (the stripe size) used by the controller. For Smart Array 6x controllers
original default stripe size for RAID 5 was 16 KB. However, the controller performed better in many
tests with a stripe size of 64 KB. In some cases, performance improved quite significantly when the
stripe size was manually changed from 16 KB to 64 KB.
As a result of this test program,
Configuration Utility (ACU) Version 7.20 or later and by HP Option ROM Configuration for Arrays
(ORCA) Version 2.54 or later. The new default stripe size is 64 KB.
HP has changed the default stripe size used by the HP Array
1
, the
1
Smart Array 6x controllers include the Smart Array 6i, 642, 6402, and 6404 controllers.
2
The test results in Tables 1 and 2 show how stripe size affected the average time to copy a 4-GB file
from Volume 1 to Volume 2 within different RAID 5 array configurations.
For the tests summarized in Table 1, the configuration was two RAID 5 volumes, each spanning three
drives. Note the significant performance improvement of the Smart Array 6i controller with the 64-KB
stripe size.
Table 1. Effect of stripe size on file copy operation (two RAID 5 volumes, three drives each)
Smart Array Controller
5i
5i
6i
6i
*Time required to copy a 4-GB file from Volume 1 to Volume 2.
Stripe size
(KB)
16 272 15.0
64 247 16.5
16 210 19.5
64 127 32.2
Time*
(s)
Rate
(MB/s)
For the tests summarized in Table 2, the configuration was two RAID 5 volumes on a single five-drive
array. In this configuration, the performance improvement with the change in stripe size was less
dramatic.
Table 2. Effect of stripe size on file copy operation (two RAID 5 volumes on a single, five-drive array)
Smart Array Controller
5i
5i
Stripe size
(KB)
16 330 12.4
64 319 12.8
Time*
(s)
Rate
(MB/s)
6i
6i
*Time required to copy a 4-GB file from Volume 1 to Volume 2.
16 276 14.8
64 228 17.9
Benchmarks used
Of the customer cases reported to HP for investigation, many involved performance of benchmarks
and utilities such as PAT, dd, and copy. These utilities do not show the true difference in performance
between the Smart Array 5i and Smart Array 6i controllers because they do not push the Smart Array
6i controller to its maximum performance. In some cases, they cannot push it even to half its maximum
performance.
As another part of this test program, HP ran a series of PAT and IOMeter performance tests. The array
configuration for this test series was one RAID 5 volume spanning three drives. The IOMeter
configuration was for 64-KB sequential reads and a queue depth of 16.
3
Table 3 summarizes the test results. In the first two rows of the table, note that with a stripe size of
16 KB, the Smart Array 5i and 6i controllers achieved identical results for the PAT utility. In the last
two rows of the table, note that in the IOMeter tests, the Smart Array 6i controller achieved a
significantly higher benchmark data rate.
Table 3. Comparison of benchmark data rates (RAID 5 volume spanning three drives)
Smart Array Controller
5i
6i
5i
6i
* IOMeter configuration: 64 KB sequential read; queue depth 16
Stripe size
(KB)
16 PAT 81
16 PAT 81
64 IOMeter* 94
64 IOMeter* 148
Utility
Rate
(MB/s)
Bus speed
The Smart Array 6i controller is an Ultra320 intelligent array controller. While many applications may
benefit from Ultra320 transfer rates, others may not. Simply stated, if the Ultra3 bus speed does not
create a performance bottleneck, then upgrading to an Ultra320 controller is unlikely to increase
performance. Configurations using a small number of drives
performance through an Ultra320 upgrade. However, performance-sensitive applications using a
small number of drives may benefit from increasing the drive count in the RAID volume to better use
the bandwidth available on the U320 bus.
2
are least likely to achieve improved
Recommendations
To maximize performance of existing Smart Array 6x controllers, HP recommends that system
administrators configure RAID volumes with a 64-KB stripe size. Changing the default stripe size is
simple using the ACU.
Note
The default stripe size for Smart Array 5x controllers is 16 KB. Therefore,
when moving a data set from a system with a Smart Array 5i controller to a
system with a Smart Array 6i controller, it is particularly important to reset
the stripe size on the Smart Array 6x controller to 64 KB.
Before upgrading to Smart Array 6x controllers for the purpose of increasing performance, HP
recommends that system administrators carefully evaluate the effect of bus speed on storage system
performance. If the existing Ultra3 bus speed does not create a performance bottleneck, then
upgrading to an Ultra320 Smart Array 6x controller is unlikely to increase performance. Instead, for
performance-sensitive applications using a small number of drives, HP recommends adding drives to
the RAID volume to better use the bandwidth available on the Ultra320 bus.
2
In this context, the definition of small depends on factors such as RAID level, drive model, and I/O profile.
4
Before upgrading existing Smart Array 5x controllers, HP recommends that system administrators
consider the characteristics of their applications. Performance benefits of upgrading from Smart Array
5x to Smart Array 6x controllers will vary by application and server configuration. Applications likely
to derive significant performance benefit from such an upgrade include imaging, data capture, and
streaming media. However, upgrading from Smart Array 5x to a Smart Array 6x controllers may or
may not produce performance benefits for the following applications: database server, web server,
and file server.
To gain an accurate indication of true performance, it is imperative to use benchmarks that match the
intended use of the hardware and software. To judge performance of the Smart Array 6i controller,
benchmarks used must duplicate the intended use of the storage subsystem.
5
For more information
Resource description Web address
ProLiant RAID controllers www.hp.com/products/smartarray
ProLiant benchmarks http://h18004.www1.hp.com/products/servers/benchmarks/index.html
Industry-Standard Server Technology
Papers
www.hp.com/servers/technology
Call to action
Please send comments about this paper to: TechCom@HP.com.
© 2005 Hewlett-Packard Development Company, L.P. The information
contained herein is subject to change without notice. The only warranties for
HP products and services are set forth in the express warranty statements
accompanying such products and services. Nothing herein should be construed
as constituting an additional warranty. HP shall not be liable for technical or
editorial errors or omissions contained herein.
TC050407BP, 04/2005
6
Loading...