Reliability ProLiant DL380 G3 servers have the proven reliability that is essential for business-critical applications. The following options are available: Redundant hot-pluggable power supplies Hot-pluggable drives Two Gigabit Network Interface Cards (NICs) Redundant system fans Up to 6GB of two-way interleaved 200MHz DDR SDRAM, with advanced ECC and online spare memory capabilities Embedded Smart Array 5i Plus controller, with optional battery-backed write cache enabler
Versatility ProLiant DL380 G3 servers are flexible, with features that include three I/O expansion slots, removable media bays, internal and external cages for hard drives, and optional DVD-ROM drive and DC power supplies. Manageability ProLiant DL380 G3 servers can improve productivity, uptime and implementation time while reducing the cost of ownership. Manageability features include: Tool-free in-the-rack access to major components Single thumbscrew system board removal Server health LED displays Front panel for power, drive and NIC status indications Pre-failure warranty Embedded Pre-boot Execution Environment (PXE) capability, allowing the server to download and execute a boot image from the network Integrated Lights-Out (iLO) capability, embedding HP's award-winning Remote Insight Lights-Out Edition on the system board to free up an expansion slot Easy-to-use system tools such as ProLiant Essentials Rapid Deployment Pack, SmartStart, SmartStart Scripting Toolkit, Insight Manager, Rack Builder Pro, ActiveUpdate, and ActiveAnswers

features

The high-performance ProLiant DL380 G3 2.8GHz server architecture includes the following features: Ultra-thin 2U chassis Intel Xeon Prestonia processor technology at 2.8GHz 512KB level-three cache 400MHz GTL bus Up to 6GB of two-way interleaved 200MHz DDR SDRAM, with advanced ECC and online spare memory capabilities Five-peer PCI bus architecture Two 64-bit/100MHz hot-plug PCI-X slots, one 64-bit/133MHz non-hot-plug PCI-X slot Embedded Smart Array 5i Plus controller, with optional battery-backed write cache enabler Support for six one-inch Wide-Ultra SCSI-2 or SCSI-3 hot-plug disk drives Up to 436.8GB of hot-plug storage (six one-inch 72.8GB drives) Two Broadcom NC7781 Gigabit Ethernet NICs with Wake-on-LAN (WOL)1 and Pre-boot Execution Environment (PXE)2 capabilities Embedded Integrated Light-Out (iLO) capability Optional DVD support Automatic Server Recovery-2 HP's industry-leading pre-failure warranty
WOL allows the administrator to remotely power on a computer or to wake it up from sleep mode. PXE allows the server to download and execute a boot image from the network.

new processor technologies
The Xeon processor used in the ProLiant DL380 G3 2.8GHz server features the key innovations and improvements described in Table 1.
Table 1. New Processor Technologies New Technology Description
Hyper-threading technology
Processor chips often use parallelism to improve performance; for example, Intels Itanium architecture is based on Explicitly Parallel Instruction Computing (EPIC) technology, which features instruction-level parallelism. However, Intels new hyper-threading technology delivers two logical processors that can simultaneously execute different tasks using shared hardware resources, improving performance by up to 40%. With instruction level-parallelism, some hardware resources are mutuallyexclusive; however, with hyper-threading, two processors share the resources of a single physical processor so that applications or threads can execute simultaneously. A single physical processor appears as two logical processors to the operating system or application. Hyper-threading can boost the performance of multi-threading and multitasking operations, and will enable the worlds first Simultaneous Multithreaded (SMT) processor. Hyper-threading delivers the following benefits to the customer: Improved performance Improved response times More users supported
More transactions processed
Hyper-threading is largely invisible to the platform. Todays multi-processoraware software is compatible with a hyper-threading-enabled platform; only small changes are needed to optimize many multi-threaded applications for hyper-threading. However, further performance gains can be realized by specifically tuning for hyper-threading. Rapid execution engine Two Arithmetic Logic Units (ALUs) on the Xeon processor run at twice the core processor frequency, allowing basic integer instructions (such as Add, Subtract, Logical AND, and Logical OR) to execute in one-half of a clock cycle.

continued

Table 1. New Processor Technologies (continued)

400MHz system bus

The Xeon processor features a 400MHz system bus with throughput capabilities up to 3.2GB per second (as opposed to 1.06GB per second delivered by the Pentium IIIs 133MHz system bus). 400MHz capability is enabled by: Physical signaling scheme The processor quad-pumps data transfers over a 100MHz clocked system bus. Buffering scheme The processor supports sustained 400MHz data transfers.
Hyper-pipelined technology
Hyper-pipelined technology can significantly increase the performance, frequency, and scalability of the processor. The pipeline depth of the Xeon processor is double that of the Pentium III; for example, the branch prediction/recovery pipeline is now implemented in 20 rather than 10 stages.
performance testing overview
HP tested one-way and two-way ProLiant DL380 G3 2.8GHz servers in a TCSC environment to determine the following critical performance characteristics: Performance HP has made a distinction between the optimal number3 of users that can be supported by a particular server configuration and the maximum number. The optimal number is a more practical metric that reflects the number of users that can be supported without impacting response times. The performance testing used various metrics to establish the optimal number of users for ProLiant DL380 G3 2.8GHz servers. System Paged Table Entry/Paged Address Pool limitation The performance testing was also intended to determine if lack of System Paged Table Entry (PTE) or Paged Address Pool space could limit server performance when there were no other system bottlenecks.
Historically, HP has defined the optimal number of users as the number of users that are active when processor utilization reaches 80%. Additional users are supported but response times may become unacceptable.

performance metrics

Performance metrics monitored by HP included the following: Processor utilization to establish the number of users at various levels of processor utilization Processor queue length to establish the lengths of the processor queues at various levels of user activity Canary time to establish the number of users that are active when response times exceed the baseline by over 10%

Note: HP used the Windows NT Performance Monitor analysis tool to monitor processor utilization (%CPU Utilization) and processor queue length (Processor Queue Length) metrics. Because users share the Terminal Servers processors to run applications, the speed and number of processors are critical for maximizing server performance. Though Terminal Services divides processing type equally between each user, processor-intensive applications may degrade overall performance.

tested configuration

This section provides information on the following topics: Server configurations An overview of the servers used in the tested configuration Recommendations Recommendations for configuring servers System summary A system summary of the server under test Client machines An overview of the clients used in the tested configuration Test scripts An overview of the test scripts used by HP to simulate the activities of Heavy, Medium and Light Users
Figure 2 illustrates the test environment.
Figure 2: The tested environment
Users accessed applications running on the ProLiant DL380 G3 server from HP Evo client machines with various configurations. HP varied the numbers of users and the levels of complexity of the work being carried out.

server configurations

Table 2 summarizes the configurations of servers deployed in the test environment.
Table 2. Server Configurations
Server Software Configuration Hardware Configuration

Terminal Services

Windows 2000 Advanced Server, Terminal Services enabled (see IMPORTANT below) Citrix MetaFrame XP
One ProLiant DL380 G3 server with: One-way or two-way 2.4GHz Xeon processor configuration 512KB integrated level-three cache 4GB RAM Two hot-pluggable 9.1GB 10,000RPM Ultra SCSI-2 hard drives Integrated Smart Array 5i+ controller with RAID 1 NC7770 PCI-X Gigabit NIC Two ProLiant DL380 G2 servers, each with: Two-way 1.266GHz Pentium III processor configuration 512KB level-two cache 2GB RAM Two hot-pluggable 9.1GB 10,000RPM Ultra SCSI-2 hard drives Integrated Smart Array 5i controller with RAID 1 Netelligent 10/100 TX PCI UTP controller

Exchange Server

Microsoft Exchange
Internet Information Server
Microsoft Internet Information Server 4.0
Two ProLiant DL380 G2 servers, each with: Two-way 1.266GHz Pentium III processor configuration 512KB level-two cache 2GB RAM Two hot-pluggable 9.1GB 10,000RPM Ultra SCSI-2 hard drives Integrated Smart Array 5i controller with RAID 1 Netelligent 10/100 TX PCI UTP controller
Table 2. Server Configurations (continued)

Database Server

Microsoft SQL Server 6.5
Two ProLiant DL580 Servers, each with: Four-way 700MHz Pentium III processor configuration 512KB level-two cache 512MB RAM Four internal hot-pluggable 9.1GB Ultra3 hard drives ProLiant storage system with six 9.1GB hard drives Integrated Smart Array controller with RAID 1 Netelligent 10/100 TX PCI UTP controller

Domain Controller

Two ProLiant DL580 Servers, each with: Two-way 700MHz Pentium III processor configuration 512KB level-two cache 512MB RAM Integrated Smart Array controller with RAID 1 Four internal hot-pluggable 9.1GB Ultra3 hard drives; RAID 1 Netelligent 10/100 TX PCI UTP controller
IMPORTANT: With users logged on, a Terminal Services server may intermittently stop responding for up to 30 seconds. HP has had limited experience with this issue but there have been numerous reports from other end-users. Microsoft has responded to this issue in Windows Service Pack 3 (SP3) but an additional hotfix is still required. For more information, refer to the Microsoft website. The customer may not achieve the test results detailed in this Performance Brief unless the Terminal Services server is running SP3 with the hotfix.

recommendations

HP offers the following recommendations for configuring servers in the test environment: Memory All applications reside and are executed on the server running Terminal Services. Adequate memory is essential to optimize system performance. HP makes the following recommendations based on user type: 5 6MB for each active Light User 8 10MB for each active Medium User 12MB for each active Heavy User
For more information on these user types, refer to the section on Test Scripts. NIC speed Since network traffic is light (only keystrokes, mouse clicks and display updates), HP set all server NICs to 100Mbps.

system summary

Table 3 provides a system summary of the ProLiant DL380 G3 server under test.
Table 3. System Summary System Component Description
Operating system Version System name System model System type Processors BIOS version Windows directory System directory Boot device Locale User name Time zone Total physical memory Available physical memory

Microsoft Windows 2000 Advanced Server 5.0.2195 Build 2195 5.0.2195 Service Pack 3 Build 2195 Administrator4 HP ProLiant DL380 G3 X86-based PC X86 Family 15 Model 2 Stepping 7 GenuineIntel~2785MHz 08/19/2002 C:\WINNT C:\WINNT\System32 \Device\Harddisk0\Partition1 United States \Administrator Central Daylight Time 3,898,900KB 3,036,224KB

client machines

HP simulated a TCSC environment using a variety of Pentium-based 800MHz 1GHz Evo workstations as clients. Each Evo was equipped with at least 256MB of memory and operated at 640480/256-color resolution. HP set all client NICs to 10Mbps.

test scripts

HP used test scripts that simulated the activities of three types of users (Heavy, Medium and Light). Table 4 describes these users and their simulated activities.
Table 4. Simulated User Activities User Type Activities

Heavy user

Heavy Users (also known as Structured Task Workers) tend to open multiple applications simultaneously and remain active for long periods of time. Heavy Users often leave applications open when not in use. Heavy users have an average typing speed of words per minute. The Heavy User script included the following activities: Creating, saving and printing documents using Microsoft Word Browsing Web pages, and sending and receiving e-mail using Microsoft Outlook Performing extensive Microsoft Excel activity (using PivotTable dynamic views, graphing and charting data, and so on)

Medium user

Medium Users (also known as Knowledge Workers) tend to open and close applications more quickly than Heavy Users. Medium Users typically do not leave applications open when not in use. Medium Users have an average typing speed of words per minute. The Medium User script included the following activities: Creating small documents with Microsoft Word Performing simple data entry in Microsoft Excel Sending and receiving e-mail with Microsoft Outlook (seven e-mail messages per hour) Engaging in minimal browsing of an intranet site using Microsoft Internet Explorer 5.5

Light user

Light Users are also known as Task Workers. The Light User script simulated a single line-of-business application a call-center application that started a Visual Basic program to access an SQL database. The script queried the database, read, wrote, and printed information.

performance test results

This section provides information on the following topics: Hyper-threading ON Projects the optimal numbers of users supported by one-way and twoway servers with hyper-threading capability turned on Performance summary Compares the performance of one-way and two-way servers Performance comparison Compares the optimal numbers of users supported by one-way and two-way servers Individual performance metrics Uses individual performance metrics to determine the maximum number of Heavy Users supported by a two-way server Canary execution time Uses an additional performance metric to validate results obtained from other performance tests System PTE/Paged Address Pool limitation shows how server performance was limited by lack of Paged Address Pool space
Note: HP determined that there were no disk, memory or network bottlenecks in the test environment.
results hyper-threading ON
HP tested one-way and two-way ProLiant DL380 G3 servers with hyper-threading ON.

One-way server

HP determined the optimal numbers of users (as shown in Figure 3) supported by a one-way ProLiant DL380 G3 server with hyper-threading ON.
Figure 3: One-way server results hyper-threading ON
As shown in Figure 3, a one-way ProLiant DL380 G3 server with hyper-threading ON can support up to 90 Light Users, 79 Medium Users, or 71 Heavy Users before the users begin to experience unacceptable response times.

Two-way server

HP determined the optimal numbers of users (as shown in Figure 4) supported by a two-way ProLiant DL380 G3 server with hyper-threading ON.
Figure 4: Two-way server results hyper-threading ON
As shown in Figure 4, a two-way ProLiant DL380 G3 server with hyper-threading ON can support up to 148 Light Users, 134 Medium Users, or 122 Heavy Users before the users begin to experience unacceptable response times.

performance summary

Table 5 summarizes performance test results with hyper-threading ON. IMPORTANT: In a different TCSC environment, the performance of a ProLiant DL380 G3 server may not match the results summarized in Table 5. To identify server configurations for alternate applications, please use the online sizer tool.

Table 5. Performance Summary with Hyper-threading ON
Processor Type Number Speed Memory Optimal Number of Users Heavy Medium Light

Xeon Xeon

2.8GHz 2.8GHz

4GB 4GB

71 122

79 134

90 148

performance comparison

Figure 5 compares the optimal numbers of users supported by one-way and two-way ProLiant DL380 G3 servers with hyper-threading ON.
Figure 5: Performance comparison with an indication of the System PTE/Paged Address Pool limitation
individual performance metrics
HP used individual performance metrics (such as processor utilization and processor queue length) to quantify server performance. HP varied the following characteristics to create a series of test scenarios: Server configuration one-way or two-way User type Heavy, Medium, or Light Hyper-threading ON
For each test scenario, HP began by running the appropriate script (simulating Heavy, Medium, or Light Users, as appropriate) for a group of ten users. Start times were staggered to eliminate authentication overhead. After the sessions finished, HP added ten more users, then repeated the testing. This methodology allowed HP to obtain the following performance metrics: Optimal number of users Historically, HP has defined the optimal number of users as the number of users that are active when processor utilization reaches 80%. Additional users are supported but response times may become unacceptable. Queue length G1 servers became congested with a queue length of 10 15. However, higher clocking rates, larger caches, and new processor technologies such as Hyper-threading allow G3 servers to process the queue more quickly while still maintaining optimal response times.

one-way server metrics

Figure 6 shows performance metrics for a one-way ProLiant DL380 G3 server with Heavy Users and hyper-threading ON.
Figure 6: Individual performance metrics for a one-way ProLiant DL380 G3 server with Heavy Users and hyperthreading ON
Figure 6 shows that 80% processor utilization first occurred when more than 71 Heavy Users were logged in. After the next group of 10 users was added, processor utilization consistently exceeded 80%; however, until processor utilization reached 100%, the server could support the additional users although response times might become unacceptable. In addition, Figure 6 shows that, with 71 Heavy Users, the processor queue length is 54.

two-way server

Figure 7 shows performance metrics for a two-way ProLiant DL380 G3 server with Heavy Users and hyper-threading ON.
Figure 7: Individual performance metrics for a two-way ProLiant DL380 G3 server with Heavy Users and hyperthreading ON
Figure 7 shows that 80% processor utilization first occurred when more than 122 Heavy Users were logged in. After the next group of 10 users was added, processor utilization consistently exceeded 80% but remained below 100%, indicating that the server could support the additional users although response times might become unacceptable. In addition, Figure 7 shows that, with 122 Heavy Users, the processor queue length was 16, 8 per processor. Note that HP is placing less importance on processor queue length as a server performance metric.

canary execution time

HP often uses canary execution time as a performance metric. With no load on the system, the canary script is executed and the time taken to execute the canary is recorded to establish a baseline. As each group of users is added and a steady state achieved, the canary is executed again. Canary scripts are executed until execution time has increased by more than 10% over the baseline. Table 6 lists canary execution times for a two-way server with Heavy Users.
Table 6. Canary Execution Times Elapsed Time Number of Execution Heavy Users Time
0:00 0:15 0:29 0:44 0:58 1:13 1:27 1:42 1:56 2:11 2:25 2:40 2:54 3:09 3:23 3:38 3:52
10:13 10:06 10:09 10:09 10:09 10:10 10:10 10:10 10:08 10:10 10:13 10:13 10:13 10:54 11:36 12:08 19:43
Table 6 indicates that there were more than 130 users when the canary execution time exceeded the baseline by 10%. This supports the figure of 122 users obtained using 80% processor utilization as a performance threshold.
System PTE/Paged Address Pool limitation
The 32-bit Windows 2000 operating system can directly address up to 4GB of address space. By default, 2GB of this space is allocated to processes, 2GB to the kernel. The kernel area includes the following areas: System Paged Table Entry (PTE) area kernel stack allocations Paged Address Pool memory allocations for users
If available memory space becomes low in one of the areas, the system may exhibit unexpected errors or the inability to accept new logins, effectively limiting system performance even though there are no other resource limitations. Figure 8 shows performance test results for Medium Users on a two-way ProLiant DL380 G3 2.8GHz server, demonstrating a performance limitation due to System PTE/Paged Address Pool space becoming low.
Figure 8: Test results showing the impact of low System PTE or Paged Address Pool space
Figure 8 shows that, in this test scenario, the server was able to support a maximum of 175 users; no additional users were able to log in. The performance limitation was not caused by processor or physical memory bottlenecks; in this case, the System PTE/Paged Address Pool space was almost consumed (as demonstrated by the Free System Page Table Entries value leveling out at 32KB), which imposed a limit on server performance. At this time, approximately 30% of processor power and 40% of the 4GB memory array were still available for use.
HP was able to determine that performance was limited by a lack of Paged Address Pool space (rather than System PTE space); however, this conclusion could not be supported explicitly using Performance Monitor. The current version of the operating system, by default, sizes System PTE to be as large as possible4, maximizing the space available to the kernel. While it is possible to reconfigure System PTE space to increase or decrease the space available to the Paged Address Pool, HP does not recommend this. Reducing the space available to the kernel may make the system unstable, outweighing the benefits of additional user support. More information on tuning System PTE and Paged Address Pool resources is available on the Microsoft website.

sizing recommendations

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 the following factors: 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 Citrix MetaFrame XP server farms inherent ability to scale out. User profiles For accurate sizing, the system architect must closely match the profiles of the users in the production environment with those described in Table 4. Simulated User Activities. If the profiles do not match, more are available from the TCSC online sizer; alternatively, the system architect can consult Compaq Global Services for more information.
On a server with Terminal Services enabled
System PTE/Paged Address Pool considerations
There is a new sizing consideration with ProLiant DL380 G3 2.8GHz servers, which are susceptible to performance limitations caused by lack of System PTE or Paged Address Pool space. The customer should size the server to accommodate the desired workload while ensuring that there will always be adequate System PTE or Paged Address Pool space available. HP recommends allowing a 25% buffer between the desired number of users and the number of users supported when the performance limitation occurs. This buffer should ensure that there is always sufficient System PTE or Paged Address Pool space available to accommodate the workload. In addition, HP recommends considering the impact of workload on available System PTE or Paged Address Pool space: Heavier workload Performance on a server with a heavy workload tends to be limited by processor power rather than lack of System PTE or Paged Address Pool space particularly since customers tend to deploy additional memory for heavier users (or reduce the number of users) Light-to-medium workload Performance on a server with a light-to-medium workload is more likely to be limited by a lack of System PTE or Paged Address Pool space since operations are less processor-intensive and there tend to be more users.

more information

A summary of the features of Terminal Services More about Terminal Services, including system architecture, client licensing, remote administration, and optimizing applications for use with Terminal Services Links to additional technical information on Windows 2000
appendix b Citrix MetaFrame XP
Citrix MetaFrame XP for Microsoft Windows offers a complete solution for organizations that want more from their application serving environments. IT administrators need more reach to connect with users in any location, on any device, over any network, and more power to manage applications from a single, centralized location. Based on proven Citrix technology, Citrix MetaFrame XP extends the reach of the Windows 2000 Server family as well as advanced platforms. It securely deploys Windows-based applications with the highest levels of manageability so administrators can easily scale servers as organizations grow. Citrix MetaFrame XP offers a more predictable, cost-effective computing experience for all users. Single-point control over applications, servers and server farms, licenses and resources means better utilization of limited IT departments.

MetaFrame XP family

Citrix specifically customized the MetaFrame XP family into three tailored solutions that accommodate a wide spectrum of customer requirements for functionality and scale: MetaFrame XPs: Delivers the level of control that departmental workgroups require to extend the reach of Windows 2000 servers to multiple devices MetaFrame XPa: Maximizes the availability of applications across the Internet for growing organizations MetaFrame XPe: Offers extensive scalability and manageability, as well as rapid application deployment, for global enterprises in a Windows 2000 server-based environment.
The MetaFrame XP family features a highly scalable communications and management foundation that provides the capability to manage and deploy applications across the enterprise. MetaFrame XP delivers much more than application support. It gives IT managers the ability and confidence to maintain a fast, predictable, and efficient application-serving network.

key benefits

Unparalleled manageability and scale systems, applications and users MetaFrame XP offers the power to manage servers and server farms anywhere across an enterprise, with robust system management capabilities that can be integrated with third-party network management solutions. At last IT managers can centrally configure and manage software distribution to multiple servers in numerous locations in just minutes, as well as monitor and control application usage, resource utilization and user activity. With these capabilities, IT managers can proactively predict system requirements and maximize availability and performance across the network. Total Net leverage integration, personalization and control The adoption of the Internet is the driving force behind the concept of a universal Net the seamless integration of LANs, WANs, the Internet, intranets and extranets that enables organizations to optimize their communications. Citrix NFuse empowers businesses with the ability to deliver any application to a single portal view without rewriting code. This is the fastest way to provide single-point access to applications and content information through any device that supports a standard Web browser improving productivity and delivering true user mobility. Ultimate flexibility any application, any device, over any Net connection MetaFrame XP provides the power to deploy any application to any device via any Net connection quickly and with a reduced total cost of ownership. By shifting application processing to the server, administrators can ensure the rapid, user-friendly deployment of business-critical applications with a higher level of performance predictability. The flexibility of seamless application access from PCs, Macintosh computers, UNIX or Linux workstations, laptops, wireless devices and other network appliances means fewer hardware roadblocks and far less technology churn. It affords a consistent user experience across the enterprise, complete freedom and mobility, unparalleled speed, and simplified management. MetaFrame XP is the powerful, behind-the-scenes framework that makes it all happen.

For more information on MetaFrame XP, please refer to the Citrix website.

HP ProLiant DL380 G3 Server Benchmark Results for Microsoft Exchange 2003
Abstract.....2 Executive Summary....2 ProLiant DL380 G3 the server behind the outstanding performance..3
Overview..... 3 Whats New..... 3 Better performance..... 3 Superior remote management..... 4 Easier ownership.... 4
Smart Array 6402 Controllers....4
Description..... 4 Features..... 5
Test methodology and workload....6 Test result highlights.....7 Appendix A Detailed test results...8
Descriptive terms..... 8 Test results...... 9 Response times.....11 Message throughput.....11 Exchange Server configuration....12 Load Generator configuration....13
Appendix B LoadSim changes from previous version..14 Appendix C MMB2 and MMB3 Workload Comparison..15 For more information.....23

Abstract

In November 2003, HP produced new Microsoft Exchange MAPI Messaging Benchmark (MMB3) results using Microsoft Windows 2003 and Microsoft Exchange 2003, on an HP ProLiant DL380 G3 server powered by two Intel Xeon 3.06 GHz processors with 512KB L2 cache. The MMB3 benchmarking workload and methodology serves as the standard for Exchange 2003 MAPI server comparison. The MMB3 workload is characteristic of a medium corporate mail environment. Using the Microsoft LoadSim utility, the HP ProLiant DL380 G3 server was tested at the HP Performance Center in Nashua, New Hampshire.

Executive Summary

HP achieved world-class Microsoft Exchange 2003 Server scalability results of 4,000 MMB3 on an HP ProLiant DL380 G3 server equipped with:
Two Intel Xeon 3.06 GHz processors with 512KB L2 cache Three Smart Array 6402 controllers for databases and transaction log 84 x 36.4GB hard disk drives for the Exchange information store and log files
The HP ProLiant DL380 G3 (dual processor) server achieved the following:
Average CPU utilization rate of 71% during the 4,000 MMB3 test. Weighted 95th percentile response-time score was 371 milliseconds. Average send-queue size for the four-hour steady-state period was 21

messages.

ProLiant DL380 G3 the server behind the outstanding performance

Overview

The HP ProLiant DL380 G3 server, as shown in Figure 1, is the industry's most versatile 2-way rack-sever, offering Integrated Lights-Out (iLO) management, the latest performance technologies, and high-availability features optimized for large data center deployments.
Figure 1. HP ProLiant DL380 G3 server
The HP ProLiant DL380 G3 server is the next generation of the award-winning dense 2-way server line and combines Integrated Lights-Out with the next generation of performance technologies to give customers more control and performance in the same space-saving form factor. Features that make this a server that will meet a wide range of deployment needs are:
Standard simplex/duplex backplane Six drive bays Three expansion slots (two of which are hot-pluggable) Optional redundant power supply and cooling fan Available DC power supplies and internal tape drive The most robust software support in the industry

Whats New

Better performance
Up to 2 Intel Xeon 3.2 GHz processors with a 1 MB L3 cache in addition to the

512K L2 Cache

ServerWorks GC-LE chipset, supporting a 533MHz Front Side Bus Three full-length PCI-X expansion slots: two hot plug 100MHz and one 133MHz 1 GB of 2-way interleaved PC2100 DDR SDRAM with Advanced ECC and Online
Spare Memory capabilities, expandable to 12 GB
HP ProLiant DL380 G3 Server Benchmark Results for Microsoft Exchange 2003 Two embedded NC7781 Gigabit Ethernet NIC ports Embedded Wide Ultra3 Smart Array 5i Plus RAID controller Single/Dual channel, enabling the 6 internal SCSI drives to run on 1 or 2 channels Optional transportable Battery-Backed Write Cache enabler for Smart Array 5i

Plus RAID controller

Superior remote management
Integrated Lights-Out (iLO) integrates the robust capabilities of the award-winning Remote Insight Lights-Out Edition without the need to consume a slot.

Easier ownership

The HP ProLiant Essentials Foundation Pack standard, including the latest SmartStart deployment software and Insight Manager 7 manageability, along with the optional Rapid Deployment Pack and iLO Advanced Pack, offer customers the most robust system-level deployment and maintenance software in the industry.
Smart Array 6402 Controllers

Description

The Smart Array 6402 high performance Ultra320, PCI-X array controller, as shown in Figure 2, provides maximum performance, flexibility, and reliable data protection for HP ProLiant servers, through its unique modular design and support for Advanced Data Guarding (RAID ADG). This new generation performance Smart Array controller again raises the standards of performance, introducing Double Data Rate (DDR) battery-backed write cache (BBWC) architecture and a new RAID engine. Designed and tested with industry standard ProLiant Servers for greater reliability, this controller is ideal for workgroup and departmental servers. And, like other Smart Array controllers, the SA-6402 features complete data compatibility with previous generations Smart Array controllers for easy data migration from server to server and for controller upgradeability.

Figure 2. Smart Array 6402 controller

Features

Modular, easy to upgrade design lets you optimize performance and increase capacity as needed from 128 to 256 MB BBWC, with future support for a two channel U320 expansion option. High performance, sixth-generation architecture offers a new hardware RAID engine and a new performance 266 MHz DDR memory architecture for increased performance over previous controllers. Recovery ROM protects against a ROM failure or corruption. Ultra320 SCSI technology delivers high performance and data bandwidth up to 320 MB/s bandwidth per channel. Mix and match LVD SCSI compatibility protects your investments and lets you deploy drives as needed. Battery-backed Cache protects cached data in the event of a power outage, server failure or controller failure, and redundant, replaceable batteries take that protection even further. Maximum cache configuration is 256 MB of battery-backed cache. 64 bit, 133 MHz PCI-X interface boosts bandwidth above 1GB/s burst transfer rate over PCI-X bus. 64 bit memory addressing supports servers with greater than 4 GB of memory. Online Management Features: Online Capacity Expansion, Online RAID Level Migration, Online Stripe Size Migration, Multiple Online Spares (Global), User Selectable Read/Write cache, User Selectable Expand and Rebuild Priority.
Test methodology and workload
For Microsoft Exchange 2000 Server, the benchmarks were measured using the MAPI Messaging Benchmark 2 (MMB2). The MMB3 workload, for LoadSim 2003, is a modification of the previous MMB2 workload. It is designed to include new features from Microsoft Exchange 2003 Server and Outlook 2003. This workload achieves the following:
Uses the Microsoft Outlook 2003 client. Introduces Smart Folders. Introduces the use of server-side rules. The message distribution is composed of bigger messages then MMB2. The mailbox size is increased to 100MB per user. Removes journaling from the benchmark.
For more detailed information regarding the differences between MMB2 and MMB3, please refer to Appendix C MMB2 and MMB3 Workload Comparison in this document. This test measures the messaging throughput of a single server, single site topology. Its purpose is to measure the maximum throughput of a Microsoft Exchange Server on this hardware configuration. Note: This test can provide a benchmark for comparing hardware and/or software products, but cannot be used as a deployment guide for production environments. For deployment specific information contact an HP representative. The MMB3 benchmark does not account for:
Usage profiles not matching that of the Load Simulator MMB3 profile; Per user storage, and per server backup requirements; Fault tolerance requirements, such as protected storage (RAID 0+1, RAID 5, etc.)

for the system/page file volume, information store and transaction logs;
Anti-virus and management processes and effects on the server; UBE/UCE (spam) mail flow; Workloads other than MAPI private folder access. This includes Public Folder,
NNTP, POP3 and other email interfaces;
Multiple Exchange Server deployments, where additional resources are required
to forward mail intra-site;
Connectors, Links and Replication to remote Exchange sites; Network topologies, bandwidth availability, latency requirement and SLA
related factors like QOS (Quality of Service) and fail-over path issues;
Manageable database sizes and partitioning beyond the 2 Storage Group, 2
database/SG configuration.

Test result highlights

Hewlett Packard Server: Test results MMB3 score: Response time: CPU utilization: Avg. queue: Msgs. submitted Msgs. delivered Msgs. sent Server configuration CPU: CPU count: RAM: Secondary cache: Operating system: Storage: 3.06-GigaHertz (GHz) Intel Xeon 2 Physical with Hyper-Threading enabled 4 gigabytes (GB) 512 Kilobytes (KB) L2 Cache Microsoft Windows 2003 Advanced Server SP36.4 GB Information Store and transaction and log files 6 36.4 GB Operating system, Exchange files, page file, and Active Directory Controller: NIC: 3 HP Smart Array 6402 controllers ProLiant NC7781 Gigabit Server Adapter 4,ms 71% 21 163,504 (4-hour steady state period) 407,046 (4-hour steady state period) 163,480 (4-hour steady state period) HP ProLiant DL380 G3
Note: Complete disclosure of test results can be found on the Microsoft Exchange 2003 Performance Scalability website:
http://www.microsoft.com/exchange/techinfo/planning/2000/PerfScal.asp
Appendix A Detailed test results

Descriptive terms

Messages Submitted Submitted calls made by clients. This equates to total messages sent by users. Messages Sent Messages that the Store sends to the categorizer in Inetinfo (SMTP Service in particular). Note: All messages even MAPI messages are sent to the categorizer, as this replaces the MTA for all but communication via X.400, with a Microsoft Exchange 5.5 server. Message Recipients Delivered Separate mailboxes where messages have been delivered. Message Opens/Sec Messages accessed for reading per second. Folder Opens/Sec Folders opened for browsing per second. RPC Read Bytes/Sec Bytes read from clients, sent via RPCs. RPC Write Bytes/Sec Bytes written to clients, sent via RPCs. IS Send Queue Average Length Send Queue Size is the number of messages in the private information store's send queue.

Test results

Table 1. Test results
Summary Supported Benchmark Load Benchmark Profile Protocol Length of Steady State Length of Test Category 4,000 MMB3s MAPI Messaging Benchmark 3 (MMB3) Exchange MAPI 4 Hours 8 Hours Single Server
Unless otherwise noted, values listed below are averages over entire 4-hour steady state period. Transactions in total Total Messages Submitted Total Message Recipients Delivered Total Messages Sent Ratio Message Recipients Delivered / Messages Submitted 163,504 407,046 163,480 2.49
Transaction Load (per hour) Messages Submitted / hour Message Recipients Delivered / hour Messages Sent / hour 40,876 101,762 40,870
Transaction Load (per Second) Message Opens/Sec Folder Opens/Sec RPC Read Bytes/Sec RPC Write Bytes/Sec 128,432 2,295,707
Transaction Queues IS Send Queue Average Length 21
(Table 1 continued next page)
HP ProLiant DL380 G3 Server Benchmark Results for Microsoft Exchange 2003 Table1. Test results (continued)
Processor Utilization System Processor Utilization (%) System Processor Queue Length System Context Switches/Sec Process % CPU Time - Store Process % CPU Time - Inetinfo Exchange 2003 server is also domain controller? (yes/no) Process % CPU Time LSASS (on domain controller) 71% 4 11,580 242% 7% Y 6%
Memory Utilization Available Bytes Pages/Sec Process Working Set Bytes - Store Process Virtual Bytes - Store 2.2 GB 1 1.2GB 2.0GB
Logical Drive Utilization IS Database Disk Reads/Sec IS Database Disk Writes/Sec IS Database Average Disk Queue Length IS Log Disk Reads/Sec IS Log Disk Writes/Sec IS Log Average Disk Queue Length (507,511,512,526) (185,172,167,164) (4,4,4,4) (0,0) (187,185) (0,0)

Response times

Table 2. Response times (Latencies)
Client Actions Send Read Reply Reply All Forward Move Delete Permanently Delete S+ Free/Busy Browse Calendar Make Appointment Request Meeting Create Smart Folder Delete Smart Folder Create Rule Delete Rule Apply View/Sort Weighted Total 95th Percentile Response Time (in milliseconds) 6328 371

Message throughput

Table 3. Summary of the MMB3 profile for an 8 hour day
Expected Messages Submitted/MMB3/Day Messages Delivered/MMB3/Day Average Recipients per Message 2.7 Measured 81.8 203.5 2.49

List Any Modifications to the default profile None
Exchange Server configuration
Table 4. Exchange Server configuration
Component Hardware Vendor Model Processor # of Processors Primary Cache Secondary Cache Other Cache Memory 4 GB
84 36.4 GB disks for Information Store files and Log files,
Hewlett-Packard ProLiant DL380 G3 Intel Xeon 3.06 GHz 2 with hyper-threading enabled

512KB L2 cache

Disk Subsystem
6 36.4 GB disks for operating system, Exchange system files, Microsoft Active Directory, and pagefiles 3 HP Smart Array 6402 Controllers
Disk Controllers Other Hardware Hardware Tunings Comments

Optimize for I/O

Mail Software Vendor Mail Server Build\Release Version Microsoft Corporation Exchange Server 2003 RTM Boot.ini /3GB MSExchESEParamLogBuffers = 1000 Userva=3030 Additional Software Tuning HeapDecommitFreeBlockThreshold=0x40000
OS Software Operating System\Version Service Pack\Patch Info File System Type Other Software
Windows Server 2003 Enterprise Edition Version 5.2.3790, Built 3790
HP ProLiant DL380 G3 Server Benchmark Results for Microsoft Exchange 2003 Table 4: Test results (continued) Network
Type of Network Network Speed MSL (sec) Time-Wait (sec) NC7781 Gigabit Server Adapter 1000 Mbps
Load Generator configuration
Table 5: Load Generator configuration
# of Load Generators (LG) Total # of LG processes Simulated Users/Process Model Processor # of Processors Memory Network Controller Operating System 1-100, 1-900,5 -650 Compaq Deskpro EN 900 MHz 1 256MB Compaq Netelligent (100) Windows XP Professional
Appendix B LoadSim changes from previous version
New user profile In the Test Properties dialog box of LoadSim 2003, there is a
Cached Mode profile. The Cached Mode profile is an example of which tasks should be enabled to simulate a cached mode user.
RPC/HTTP LoadSim 2003 allows simulation of RPC/HTTP deployments in its
entirety. You can use the Test/Logon tab to configure your RPC/HTTP settings, including SSL encryption. Folders, Offline Address Book, Synchronize Folders tasks to better simulate real-life users.
Outlook 2003 specific tasks In LoadSim 2003, you can use the new Smart
Dynamic Distribution Lists In Topology Properties dialog box of LoadSim 2003,
there is a Dynamic Distribution List (DDL, also known as Query-Based Distribution Group) group of settings that allow the creation of DDLs. The Send Mail task also allows sending mail to DDLs with a desired frequency.
Rules In LoadSim 2003, you can populate users with server-side rules and
simulate how users create and delete them over time. All LoadSim-created rules are visible and capable of being manipulated by Outlook.
Profile improvements The medium and heavy profiles have enhanced
simulation capabilities that are disabled by default in MMB2. These improvements include keeping messages open and loading message properties in a consistent manner with Outlook 2003.

Appendix C MMB2 and MMB3 Workload Comparison
Table 6. Topology properties
Topology properties Security Use a separate account for each Exchange user Use one account for all Exchange users Credentials Logon to users using their respective accounts Distribution Lists Use Distribution Lists Number of Distribution Lists per site Distribution List minimum/average/maximum Dynamic Distribution Lists (DDLs) Use DDLs Create one for all LoadSim Users Create one per MDB No No No No No No Yes 100 2/10/20 Yes 1000 2/10/20 No No No Yes Yes Not Tested MMB2 MMB3

Table 7. Test Properties

Test Properties Tasks Send Mail Number of times per day Priority percent high Priority percent low Request receipts percent delivery Request receipts percent read Request receipts percent both (Table 7 continued next page) MMB2 MMB3
HP ProLiant DL380 G3 Server Benchmark Results for Microsoft Exchange 2003 Table 7. Test properties (continued)
Test Properties Filename (Weight) Oups1k.msg Oups2k.msg Oups4k.msg Oups10kat.msg OupsWDatt.msg OupsXLatt.msg OupsBMobj.msg OupsXLobj.msg McPP1Matt.msg McPP100katt.msg McWD2Matt.msg Recipients per message Add a Distribution List to percent message sent Add a DDL to percent message sent Save a copy in Sent Items Process Inbox Read new mail per day Load MAPI Properties as Outlook 2003 does Apply Random views to Inbox % of the time Message Actions Reply Reply All Forward Delete Copy Move (Table 7 continued next page)
Yes No Not Tested Yes Yes 75

1 1-5, avg 0

2 1-5, avg 0
Test Properties Message Actions Permanently Delete Read note delay minimum/average/maximum Load percent of attachments Accept percent of meeting requests Open Messages Keep up to messages open after browsing Leave messages open % of the time Maximum Messages in Folders No Limit Limit number of messages Browse Mail Browse mail per day Apply Random views to Folders % of the time Open Messages Keep up to messages open after browsing Leave messages open % of the time Maximum Messages in Folders No Limit Limit number of messages Public Folder Post Browse Public Folders Free/Busy Update schedule times per day Update Free/Busy information Schedule size (KB) minimum/maximum/average (Table 7 continued next page)

4 No 5/40/Yes 5/40/22 Yes Not Tested Not Tested Yes Not Tested Not Tested Not Tested Not Tested 15 Not Tested No 125 No 1.0/1.0/1.0.0/0.0/0.70
Test Properties Request Meetings Make new meetings per day Meeting Length (in hours) minimum/average/ maximum Attendees minimum/average/maximum Add a Distribution List percent of the time Make Appointments New appointments per day
Appt length minimum/average/maximum Percent recurring appointments Percent all day events Browse Calendar Number of times per day Journal Mail Items 4 1/3/1/3/1/2/7 1/5/2 1/2/7 1/5/40 20
Number of times per day Journal Applications Activity Number of times per day Logoff Number of times per day to log off Always keep connection Empty Deleted Items Browse Contacts Number of times per day Create contact Number times/day to make new contact Smart Folders Number of times per day (Table 7 continued next page)

Not Tested

3 No Yes

Not tested

Test Properties Actions Delete Create Browse Number of Smart Folders (min/max) Criteria for mail in Smart Folders Unread Important Old Mail For follow up Conversation with random person or DL Unread or for follow up Received this week From random person or DL Large (size) Sent to random DL With specific word Rules Number of times per day Activities with Rules Delete Create Number of Rules (min/max) Conditions From DL From person Sent only to me (Table 7 continued next page) Not tested Not tested Not tested 0
Not tested Not tested Not tested 0/10 Not tested 3 Not tested Not tested Not tested Not tested Not tested Not tested Not tested Not tested Not tested Not tested Not tested 10 Not tested Not tested Not tested Not tested 0 3/10
Test Properties Conditions With word in Subject or body Name in To/CC box With attachment Name not in To Box With word in subject Actions Delete Move to Folder Copy to Folder Forward to Person Forward to DL Permanently Delete Stop processing more rules Test/Logon Logging on immediately at the very beginning of the test Logging off at the end of each simulated day Empty Deleted Items folder while logging off Test Report: Approximate Message Traffic, per User, per Day Total received Reply Reply All Forward Total submitted Average Number of Recipients per Message (All Messages) Approximate receipts requested, per user, per day Read receipts Delivery receipts
185 20.56 6.48 10.3.46.61 4.78 4.2.44
Not tested Not tested Not tested Not tested Not tested
Not tested Not tested Not tested Not tested Not tested Not tested Not tested

Yes Yes Yes

Table 8. Initialization properties

Initialization Properties Mailbox Setup Number of messages in Inbox Number of messages in Deleted Items Number of new folders Messages per new folder Number of smart folders Number of rules in inbox Initialize Free/Busy Information Calendar Setup Number of appointments Contacts Setup Number of contacts

Table 9. Action Weights

Action Weights (for calculating response times) Mailbox Setup Send Read Reply Reply All Forward Move Delete Permanently Delete S+ Free/Busy (Table 9 continued next page) MMB2 MMB3
HP ProLiant DL380 G3 Server Benchmark Results for Microsoft Exchange 2003 Table 9. Test properties (continued)
Browse Calendar Make Appointment Request Meeting Create Smart Folder Delete Smart Folder Create Rule Delete Rule Apply View/Sort 1 1
All messages were converted to have HTML bodies in LoadSim 2003.

For more information

The following key documents and locations provide a wealth of information regarding successful deployment of Microsoft Exchange Server on HP platforms. ActiveAnswers http://www.hp.com/solutions/activeanswers Managing and Monitoring Microsoft Exchange Server Microsoft Exchange Server Performance and Configuration Guide Implementing High Availability for Microsoft Exchange Server Microsoft Exchange Server http://www.microsoft.com/exchange
2004 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. Microsoft is a U.S. trademark of Microsoft Corporation. Xeon is a trademark or registered trademark of Intel Corporation or its subsidiaries in the U.S. and other countries. 1/2004