For more information.......................................................................................................................... 19
Executive summary
Important:
This document describes a performance characterization performed
utilizing the HP 64-bit test harness, which incorporates a Microsoft® Office
2003 workload.
Test results cannot be compared directly with the results of tests performed
using the 32-bit Office XP- or Office 2003-based harness.
The HP ProLiant DL585 G5 is a highly manageable, rack-optimized, four-socket1 server designed for
maximum performance in an industry-standard architecture. With up to four Quad-Core AMD
Opteron™ processors and a large memory footprint, the HP ProLiant DL585 G5 server delivers the
performance and performance-per-watt needed for compute-hungry database, virtualization, and
consolidation applications. Its impressive remote management capabilities can help reduce your costs
and improve your ability to respond quickly to business changes.
Key benefits include:
• Outstanding performance and reliability
• Industry-leading management options
• Flexibility for a variety of enterprise deployments
• High performance-per-watt
A 4P HP ProLiant DL585 G5 server equipped with the Quad-Core AMD Opteron Processor Model
8360 SE (2.5 GHz) can provide optimal support for up to the following numbers of users (as
described in
Table 2) in a 64-bit HP Server Based Computing (SBC) environment:
Heavy Users
Medium Users
Light Users
288
444
528
Since the kernel memory constraints that limit scalability in a 32-bit HP SBC environment have been
removed, this performance characterization demonstrates that customers can expect to fully utilize the
resources of this server in a 64-bit environment, even when running their 32-bit applications.
Target audience: This performance characterization is intended primarily for IT professionals planning
HP SBC solution deployments. The performance and sizing information provided herein is designed to
help customers estimate the number of HP ProLiant DL585 G5 servers required for a particular
environment.
Testing performed in June 2008 is described.
1
Four-processor, also known as 4P
2
Introduction
The HP ProLiant DL585 G5 server (shown in Figure 1) is ideal for multi-threaded, multi-tasked
environments, high-performance computing, and HP SBC.
Figure 1. HP ProLiant DL585 G5 server
AMD Opteron processors
The HP ProLiant DL585 G5 server supports up to four Quad-Core AMD Opteron 8300 Series
processors. This native quad-core processor delivers the following benefits:
•Outstanding performance
The Quad-Core AMD Opteron processor is designed for optimal multi-threaded application
performance. Its native quad-core implementation features four cores on a single die for more
efficient data sharing, while the enhanced cache structure and integrated memory controller can
sustain application throughput. This processor provides outstanding processing power and, together
with its performance-per-watt enhancements, can improve IT responsiveness while maintaining data
center costs.
•Enhanced power efficiency
Thanks to Enhanced AMD PowerNow!™ technology and the introduction of AMD CoolCore™
technology, Quad-Core AMD Opteron processors are very power-efficient, helping to reduce power
needs and cooling costs in the data center.
(For more on power management enhancements, see
•Optimal virtualization
Featuring AMD Virtualization™ (AMD-V™) technology with nested paging acceleration, Quad-Core
AMD Opteron processors can accelerate the performance of virtualized applications and improve
efficiency when switching between virtual machines; as a result, customers can typically host more
virtual machines and users per system, maximizing the consolidation and power-saving benefits of
virtualization.
• Investment protection
By leveraging AMD’s Common Core Strategy and Same Socket Technology, Quad-Core AMD
Opteron processors can minimize changes to the customer’s software and data center infrastructure,
protecting IT investments and simplifying management.
AMD Dual Dynamic Power Management.)
3
AMD Dual Dynamic Power Management
Without compromising performance, AMD Dual Dynamic Power Management functionality allows
each processor to maximize the power-saving benefits of Enhanced AMD PowerNow! by reducing
idle power consumption and enabling per-processor power management in multi-socket systems to
further reduce power consumption.
By powering core and memory controller voltage planes independently, AMD Dual Dynamic Power
Management can enhance both performance and power management.
Benefits include:
•Increased performance
The memory controller is able to run at a higher frequency, helping to reduce memory latency and
thus improving application performance.
•Improved power management
By operating independently from the memory controller, the cores in a Quad-Core AMD Opteron
processor can exploit the power savings offered by Enhanced AMD PowerNow! more often,
resulting in reduced power and cooling bills. In addition, the processor reduces power to the
northbridge
2
when memory is not in use, while continuing to provide full power to the cores.
The following sections of this paper describe the testing performed by HP to characterize the
performance and scalability of an HP ProLiant DL585 G5 server in a 64-bit HP SBC environment.
Note:
A 64-bit HP SBC environment eliminates the kernel memory constraints that
can limit server scalability in a 32-bit HP SBC environment. For more
information, refer to
x64 Editions
Appendix A – Using Microsoft Windows Server 2003
.
Test methodology
HP continues to upgrade existing HP ProLiant servers and introduce new servers to meet particular
business needs. To help customers select the appropriate server for their particular HP SBC
environment, HP publishes this and other performance characterizations so that you can compare
individual server performance and scalability.
This section describes how HP determined the optimal number of users supported by a 4P HP ProLiant
DL585 G5 server equipped with the Quad-Core AMD Opteron Processor Model 8360 HE
GHz) – henceforth referred to as the HP ProLiant DL585 G5 server – in a 64-bit test harness.
Important:
As with any laboratory testing, the performance metrics quoted in this
paper are idealized. In a production environment, these metrics may be
impacted by a variety of factors.
HP recommends proof-of-concept testing in a non-production environment
using the actual target application as a matter of best practice for all
application deployments. Testing the actual target application in a
test/staging environment identical to, but isolated from, the production
environment is the most effective way to characterize system behavior.
3
(2.5
2
Or memory controller hub (MCH)
3
HE processors are higher power, frequency-optimized devices.
4
This section provides more information on test tools, user profiles and test scenarios.
Test tools
To facilitate the placement and management of simulated loads on an HP SBC server, HP used
Terminal Services Scalability Planning Tools (TSScaling), a suite of tools developed by Microsoft to
help organizations with Microsoft Windows® Server 2003 Terminal Server capacity planning.
Table 1 describes these tools.
Table 1. Components of TSScaling
Component Description
Automation tools Robosrv.exe Drives the server-side of the load simulation
Robocli.exe Helps drive the client-side of the load simulation
Test tools Qidle.exe
Tbscript.exe
Help files TBScript.doc Terminal Server bench scripting documentation
TSScalingSetup.doc A scalability test environment set-up guide
TSScalingTesting.doc A testing guide
Determines if any scripts have failed and require
operator intervention
A script interpreter that helps drive the client-side load
simulation
More information
• Roboserver (Robosrv.exe) and Roboclient (Robocli.exe):
• TSScaling:
Windows Server 2003 Terminal Server Capacity and Scaling
Terminal Server capacity planning
5
User profiles
To simulate typical workloads in this environment, HP used scripts based on the Heavy, Medium, and
Light User profiles described in Table 2.
Table 2. User profiles incorporated into the test scripts
User class Activities
Heavy User
Medium User
Light User
Heavy Users (also known as Structured Task Workers) tend to open multiple applications
simultaneously and remain active for long periods. Heavy Users often leave applications open
when not in use.
Medium Users (also known as Knowledge Workers) are defined as users who gather, add value
to, and communicate information in a decision-support process. Cost of downtime is variable but
highly visible. These resources are driven by projects and ad-hoc needs towards flexible tasks.
These workers make their own decisions on what to work on and how to accomplish the task.
Sample tasks include: marketing, project management, sales, desktop publishing, decision
support, data mining, financial analysis, executive and supervisory management, design, and
authoring.
Light Users (also known as Data Entry Workers) input data into computer systems. Activities
include transcription, typing, order entry, clerical work and manufacturing.
Table 3 outlines the activities performed by each user class utilizing Office 2003 products.
Table 3. Activities incorporated into the test scripts for each user class
Activity description Heavy User Medium User Light User
Access
Open a database, apply a filter, search through
records, add records, and delete records.
X
Excel Open, print and save a large spreadsheet. X X X
Excel_2
InfoPath
Outlook
Outlook_2 Create a long reply. X
PowerPoint
PowerPoint2
Word Create, save, print, and email a document. X
Create a new spreadsheet, enter data, and create
a chart. Print and save the spreadsheet.
Enter data
existing form.
First pass: Email a short message.
Second pass: Email a reply with an attachment.
Create a new presentation, insert clipart, and
apply animation. View the presentation after each
slide is created.
Open and view a large presentation with heavy
animation and many colors and gradients.
4
into a form; save the form over an
X X
X X
X
X X
X
5
X X
4
Data entry for Office InfoPath 2003 requires significant processor resources
5
Shortened version for Heavy Users
6
Test scenarios
For the Heavy User type, HP initiated testing by running the appropriate script with a group of 15
simulated users. Start times were staggered to eliminate authentication overhead. After these sessions
finished, HP added 15 more users, then repeated the testing. Further groups of 15 users were added
until the maximum number of users was reached.
For Medium and Light User types, HP utilized groups of ten users.
Performance and scalability metrics
While the scripts were running, HP monitored a range of Windows Performance Monitor (Perfmon)
counters to help characterize server performance and scalability. In particular, HP has monitored CPU
utilization (% Processor Time) to establish the optimal number of users supported on an HP SBC server
– by definition, the number of users active when processor utilization reaches 80%. At this time, a
limited number of additional users or services can be supported; however, user response times may
become unacceptable.
To validate scalability metrics obtained using Perfmon, HP also runs canary scripts to characterize
Heavy User response times – a very practical metric – for discrete activities such as an application
being invoked or a modal box appearing. By monitoring response times as more and more users log
on, HP has been able to demonstrate that these times are acceptable when the optimal number of
users (as determined using Perfmon counter values) is active.
Note:
When running canary scripts, HP considers user response times to become
unacceptable when they increase markedly over a baseline measurement.
7
Test topology
Figure 2 illustrates the HP SBC test environment.
Figure 2. The tested environment
Note:
Test environments such as that shown in Figure 2 are available to customers
HP Solution Centers to help solve a wide variety of business problems.
at
8
Configurations
Table 4 summarizes the configurations of servers and clients used in the test environment.
Table 4. System configurations
Server Configuration
HP SBC server
Exchange Server/
Internet Information Services
4P HP ProLiant DL585 G5 server with:
• Quad-Core AMD Opteron Processor Model 8360 SE (2.5 GHz)
– 512 KB L2 cache per core; 2 MB L3 cache
• 32 GB RAM
• Smart Array P400 controller with RAID 0
• Four 72 GB 15,000 rpm SAS hard drives
– 48 GB page file on system partition
• NC371i Multifunction Gigabit Server Adapter
Windows Server 2003 R2 Enterprise x64 Edition with Service Pack 2;
Terminal Services enabled
Office 2003
2P HP ProLiant DL360 G5 server with:
• Dual-Core Intel® Xeon® processor (3.2 GHz)
• 2 x 2 MB L2 cache
• 2 GB RAM
• Four 72 GB 15,000 rpm SAS hard drives
• Integrated Smart Array P400i controller with RAID 5
• NC373i Multifunction Gigabit Server Adapter
Windows Server 2003 Enterprise Edition
Microsoft Exchange Server 2003
Microsoft Internet Information Services (IIS) 6.0
Domain Controller
2P HP ProLiant DL360 G5 server with:
• Dual-Core Xeon processor (3.2 GHz)
• 2 x 2 MB L2 cache
• 2 GB RAM
• Four 72 GB 15,000 rpm SAS hard drives
• Integrated Smart Array P400i controller with RAID 5
• NC373i Multifunction Gigabit Server Adapter
Windows Server 2003 Enterprise Edition
Continued
9
Table 4. System configurations (continued)
Server Configuration
Client
Variety of Intel Pentium®-based Compaq Evo workstations (600 MHz – 2.533 GHz),
each with:
• At least 256 MB of memory
• 1024 x 768/16-bit color depth
• 100 Mbps NIC
Windows 2000 Professional or Windows XP
HP SBC server summary
Table 5 summarizes the configuration of the tested HP SBC server.
Table 5. System summary for the HP ProLiant DL585 G5 server
Component Description
Operating system (OS) Microsoft Windows Server 2003, Enterprise x64 Edition
Version 5.2.3790 Service Pack 2, Build 3790
Other OS description R2
System name DL585
System model ProLiant DL585 G5
System type x64-based PC
Processor AMD64 Family 16 Model 2 Stepping 3 AuthenticAMD ~2512 MHz
BIOS version/date HP A07, 3/28/2008
SMBIOS version 2.4
Windows directory C:\WINDOWS
System directory C:\WINDOWS\system32
Boot device \Device\HarddiskVolume1
Locale United States
Hardware abstraction layer
version
User name [Not available]
Time zone New Zealand Standard Time
5.2.3790.3959 (srv03_sp2_rtm.070216-1710)
Continued
10
Table 5. System summary for the HP ProLiant DL585 G5 server (continued)
Component Description
Total physical memory 32,765.62 MB
Available physical memory 30.63 GB
Total virtual memory 79.06 GB
Available virtual memory 78.54 GB
Page file space 48.00 GB
Page file C:\pagefile.sys
Performance test results
This section outlines the test results used by HP to characterize the performance and scalability of the
HP ProLiant DL585 G5 server.
• Perfmon values – Shows select Perfmon counter values for the Heavy User scenario
• Canary times – Shows Heavy User response times for a sample canary script
Note:
As with any laboratory benchmark, the performance metrics quoted in this
performance brief are idealized. In a production environment, these metrics
may be impacted by a variety of factors; for more information, refer to
Appendix B – SBC solution sizing.
HP determined that there were no disk or network bottlenecks in the test
environment.
11
Perfmon values
HP ran a series of performance tests using scripts based on the Heavy, Medium, and Light User
profiles. Figure 3 shows the test results for Heavy Users.
Figure 3. % Processor Time values for Heavy Users – showing that 288 Heavy Users were supported when processor utilization
reached 80%
Figure 3 shows the optimal number of Heavy Users supported by the HP ProLiant DL585 G5 server
to be 288
.
12
Canary times
Figure 4 shows sample results for the HP ProLiant DL585 G5 server running a typical canary script.
Individual user response times are shown in blue, with a yellow line depicting average response
times.
HP analyzed this figure to determine when response times began to increase markedly and
consistently over a baseline level, indicating that user response times had become unacceptable.
Figure 4. Canary time values show that response times began to become unacceptable when 327 Heavy Users were active
Figure 4 indicates that the HP ProLiant DL585 G5 server could support 327 Heavy Users before
response times started to increase markedly, validating the optimal value of 288 Heavy Users
derived using % Processor Time values.
13
Test analysis summary
Figure 5 summarizes the optimal numbers of users supported by the HP ProLiant DL585 G5 server.
Figure 5. Optimal numbers of users supported in the 64-bit test harness
HP characterized the scalability of the HP ProLiant DL585 G5 server through the numbers of users
supported when CPU utilization reached 80%.
Recommendations
Since x64 platforms allow you to better utilize memory and multi-core processors, the bottleneck you
are most likely to encounter
6
is associated with the disk subsystem. While a detailed analysis of disk
I/O performance is beyond the scope of this white paper, the following observations are offered to
help you improve disk performance:
• Utilize battery-backed write cache (see
Using BBWC).
• Since built-in storage is often insufficient to support a large number of users in an HP SBC
environment, consider deploying additional RAID arrays/SAN support.
Note also that when a SCSI RAID array is used to host user profiles and page files, the number of
spindles deployed has a significant impact on the response times associated with file access.
• When the pressure on the disk I/O subsystem is high, one option for improving disk access times is
to add RAM to lower the pressure on memory.
As the number of sessions increases, disk activity and the pressure on the disk I/O subsystem also
increase. If file I/O activity is high, the probability that requests will find the desired data in
memory decreases, thus negatively affecting file access times.
To help avoid a disk I/O bottleneck, Microsoft recommends using the Windows performance
monitoring tool, Perfmon, to check the following metrics
7
:
• %Idle time – Idle times for logical and physical drives should average at least 50%
• Average Disk Seconds/Read and Average Disk Seconds/Write – The average time taken to
complete a read or write should average less than 25 milliseconds, with peaks less than 50
milliseconds
If the above conditions specified by Microsoft cannot be met, a disk I/O bottleneck is likely.
6
For further information, refer to the HP white paper, “Scalability and performance of HP ProLiant servers on 64-bit Microsoft
Windows Server 2003 in an HP SBC environment.”
7
For further information, visit the Microsoft website; click the Start download link.
14
Note:
In the event of an I/O bottleneck, you should tune the disk subsystem,
decrease the number of users or applications, or add memory to the server.
Using BBWC
HP Smart Array controllers include an allocation of RAM that can be utilized to temporarily buffer
data being written to or read from disk. Since access to this RAM is significantly faster than disk
access, cache can enhance overall server performance, particularly during login operations.
Write cache is of particular interest in the HP SBC environment. After buffering all the data associated
with a particular write command, the Smart Array controller indicates to the HP SBC server that the
data transfer to the disk is complete – even through the data is still being written to disk. This frees up
the server’s processor to perform other tasks and accelerates data throughput.
Performance improvements from write cache are typically most significant when the HP SBC server
is performing log-intensive operations and/or when significant page file write operations are
necessary, such as during user logins. Lab testing has demonstrated performance gains ranging
from 50% to 250%
particular HP SBC environment.
Note:
Write cache was not used during the testing described in this report.
8
; actual results will vary depending on the application(s) involved and your
Enabling write cache
While faster access times can translate directly to enhanced system performance, this improvement
comes with a potential penalty: if a system or utility failure were to occur during cache
synchronization, data might be lost unless battery back-up has been provided.
Because of the potential for data loss, write cache is automatically disabled on HP SBC servers unless
an optional Battery Backed Write Cache (BBWC)
9
enabler is installed and fully-charged. BBWC
allows you to cache write data safely, knowing that your data would be protected by the battery in
the event of a system or controller.
HP highly recommends implementing BBWC on every HP SBC server.
Summary
When planning an x64 HP SBC environment, you should select servers equipped with multi-core
processors to help maximize scalability. If your budget allows, consider the fastest processors, the
most cores, and largest cache.
Memory is an important factor in the x64 environment: while an x64 platform can utilize more RAM,
it also has a higher minimum RAM requirement than an x86 platform. Since a system that is not
memory-starved may also prevent disk I/O bottlenecks, HP recommends adding as much RAM as
your budget permits.
For optimal performance, ensure that BBWC is implemented and enabled, in conjunction with the
largest possible controller cache. Consider using 15,000 rpm SAS drives.
8
For further information, visit the HP website
9
BBWC is available as an option for select HP ProLiant servers.
15
Appendix A – Using Microsoft Windows Server 2003 x64
Editions
Microsoft offers high-performance platforms for 64-bit applications with continued support for 32-bit
applications and existing deployment and management tools – all on the same platform. These
operating systems provide an evolutionary path to 64-bit technology, allowing 64-bit and 32-bit
applications to run side-by-side during the gradual migration to 64-bit computing.
64-bit editions of Windows Server 2003 running on Quad-Core AMD Opteron processors can
improve the performance of HP SBC servers by processing more data per clock cycle, addressing
more memory, and running some numerical calculations faster. Large data sets can be loaded entirely
into memory, reducing the need for slower disk access; complex calculations that take hours to
complete on a 32-bit system can be performed in minutes; and workloads that once required a large
server farm can be performed by a single server.
In addition, this 64-bit platform also removes many of the limitations that have previously inhibited
scalability in an HP SBC environment.
Historical scalability limitations
32-bit Windows operating systems can directly address 4 GB of memory, 2 GB of which is reserved
for the operating system kernel and 2 GB for applications. Since kernel memory is shared by all
applications, the relatively small size of this space can be particularly problematic in an HP SBC
environment where a server may be responsible for hundreds of users and thousands of processes. In
this scenario, kernel memory can become constrained, making user response times unacceptably long
and effectively limiting the ability of the server to scale up.
Historically, HP SBC environments have been implemented using 1P or 2P servers. Larger, more
powerful servers have typically not been deployed for two main reasons:
• Kernel memory issues have limited the performance of more powerful servers; either a disk I/O
bottleneck occurs or kernel memory is consumed before processor resources can be fully utilized
• Scalability in a 32-bit symmetric multi-processing (SMP) system is inherently non-linear above 2P
With these 1P and 2P server farms, opportunities to scale up are limited. As a result, customers are
forced to scale out, which can create new problems such as deployment and management
complexity, high power and cooling requirements, under-utilized resources, and minimal opportunities
for server consolidation.
The 64-bit platform shatters the earlier 4 GB limitation – for example, Windows Server 2003 R2
Datacenter x64 Edition with Service Pack (SP) 1 supports up to 2 TB of RAM – effectively removing
kernel memory limitations and eliminating disk I/O bottlenecks. By deploying a Windows Server
x64 Edition operating system, you can fully utilize the resources of your existing HP SBC servers
and take full advantage of new, more powerful systems – whether you are running 32- or 64-bit
applications.
More information
For more information on the impact of 64-bit Windows Server 2003 x64 Editions in an HP SBC
environment, refer to the HP
bit Microsoft Windows Server 2003 in an HP SBC environment.”
white paper, “Scalability and performance of HP ProLiant servers on 64-
To learn about 64-bit computing in an HP SBC environment, refer to the HP
“Fundamentals of 64-bit computing in an HP SBC environment.”
white paper,
16
Appendix B – SBC solution sizing
As with any laboratory benchmark, the performance metrics quoted in this performance brief are
idealized. In a production environment, these metrics may be impacted by a variety of factors,
including the following:
•Overhead
Agents and services (virus scanning, backup and restore, provisioning, security, management and
more) automatically consume overhead. Rogue applications can consume additional overhead.
The system architect may wish to provide a 25% – 30% buffer to accommodate this overhead.
•Future growth
To accommodate future growth, the system architect may wish to provide an additional buffer.
Alternatively, servers can be added as needed, taking advantage of the server farm’s inherent
ability to scale out.
•User profiles
The particular application in use directly impacts the number of users supported by a particular
server. Further, user behavior can also impact scalability:
– Increased typing rates correspond to fewer users.
– Opening and closing applications (rather than switching between them) or moving quickly
between tasks can place a heavier load on the server.
– For accurate sizing, system architects should closely match their user profiles with the Heavy,
Medium, and Light User profiles specified by HP in
are available using the online sizer tool (described below); alternatively, the system architect
can consult
•Background grammar checking
Background grammar checking can significantly impact scalability, reducing the number of users
supported by as much as 50%. HP disabled background grammar checking for the testing
described in this performance brief.
HP Services for more information.
Table 2. If the profiles do not match, more
Online sizer tool
To minimize risk, HP offers automated, online tools that can help the customer size an HP SBC
solution. The algorithms and methodology used by the sizer are based on the results of customer
surveys and thorough testing.
HP ProLiant Sizer for Citrix XenApp and Microsoft Windows Server 2003 Terminal Services is
An
available for enterprise and small and medium business (SMB) environments. Figure B-1 shows the
Application Selection screen for this sizer.
17
Figure B-1. The sizer’s Application Selection screen
Based on information provided by the customer, the sizer can provide a quick, consistent mechanism
for identifying the “best-fit” server for a particular HP SBC environment and generate a Bill of
Materials (BOM) for that server.
18
For more information
HP ProLiant DL585 G5 server http://www.hp.com/servers/dl585
HP ActiveAnswers for Server Based
http://www.hp.com/solutions/activeanswers/hpsbc
Computing
HP ProLiant Sizer for Citrix XenApp
(formerly Presentation Server) and Microsoft
Microsoft and Windows are U.S. registered trademarks of Microsoft Corporation.
AMD Opteron, AMD PowerNow!, AMD Dual Dynamic Power Management, AMD
CoolCore, AMD Virtualization, and AMD-V are trademarks of Advanced Micro
Devices, Inc. Intel, Xeon and Pentium are trademarks of Intel Corporation in the U.S.
and other countries.
4AA1-8622ENW, June 2008
www.hp.com/solutions/feedback.
Loading...
+ hidden pages
You need points to download manuals.
1 point = 1 manual.
You can buy points or you can get point for every manual you upload.