HP D7171A - NetServer - LPr Instruction Manual

HP NetServer
Performance and Sizing
Guidelines
Microsoft Windows NT 4.0 Server,
Terminal Server Edition
January 1999

Hewlett-Packard Company 2
Notice
The information contained in this document is subject to change without notice.
Hewlett-Packard makes no warranty of any kind with regard to this material, including, but not
limited to, the implied warranties of merchantability and fitness for a particular purpose.
Hewlett-Packard shall not be liable for errors contained herein or for incidental or consequential
damages in connection with the furnishing, performance, or use of this material.
The use of trademarks or other designations is for reference purposes only. The following
trademarks are used in this manual:
Intel, Pentium and Xeon are registered trademarks of Intel Corporation. Microsoft and MS-
DOS are U.S. registered trademarks of Microsoft Corporation. Windows and Windows NT are
trademarks of Microsoft Corporation. Citrix WinFrame and ICA are registered trademarks of
Citrix Systems, Inc.
Hewlett-Packard
Network Server Division
10955 Tantau Avenue
Cupertino, California 95014
© Copyright 1999, Hewlett-Packard Company

Hewlett-Packard Company 3
Contents
2 TEST SUMMARY............................................................................4
Introduction......................................................................................................................4
Processors.........................................................................................................................5
Memory............................................................................................................................7
Network............................................................................................................................7
Client Solutions ................................................................................................................8
3 SIZING ACTIVITY..........................................................................10
Methodology...................................................................................................................10
Test Configuration .......................................................................................................10
Visual Test Scripts.......................................................................................................14
Test Procedures............................................................................................................16
Test Results ....................................................................................................................16
Elapsed Time...............................................................................................................17
Process Utilization........................................................................................................18
Network Utilization......................................................................................................24
Memory Utilization......................................................................................................26

Hewlett-Packard Company 4
2 Test Summary
Introduction
In a Microsoft® Windows NT® 4.0, Terminal Server (TSE) multiuser
environment users share hardware and software resources on the server.
These resources include processors, memory, and storage as well as operating
system resources and application resources such as configuration data (e.g.,
location of directories and files, arrangement of application within groups, and
desktop appearance).
In order to help standardize sizing information, Microsoft has defined three
categories of users:
• Task-based users. A task-based user typically runs a single application
used for data entry, for example a line-of-business application written in
Visual Basic (POS, Order Entry applications, SAP, and so on).
• Typical user. A typical user runs one or two applications but usually runs
only one at a time. A typical user runs applications whose data processing
demands on the system are not heavy, for example a productivity
application such as a word processor, a web browser or an e-mail client.
• Advanced users. An advanced user is a more sophisticated user who runs
three or more applications, and often has several applications active at the
same time. An advanced user runs applications whose data processing
demands on the system are heavy.
For the sample configurations described in the following table, Microsoft
provides, for each type of user, the recommended number of users that can be
supported (Microsoft Window NT Server, Terminal Server Administrator's
Guide).
Configuration Task-based users Typical users Advanced users
Single processor Intel®
Pentium®Pro 200MHz
128MB RAM
25 15 8
Dual-processor Pentium Pro
200MHz 256MB RAM 50 30 15
Quad-processor Pentium
Pro 200MHz 512MB RAM 100 60 30
Table 1

Hewlett-Packard Company 5
Processors
In the TSE environment, all users contend for the same processors. In this
environment a single CPU-intensive application can adversely effect the
performance of all users on the system.
As expected, our results also show that as the CPUs become saturated,
performance deteriorates significantly and client sessions can become starved.
As a result, solutions consisting of well behaved, predictable applications (e.g.,
order entry applications) are ideally suited for this environment.
The following table contains general guidelines, for the number of users that
can be supported by on a sample set of HP NetServer configurations.
HP NetServer
Configuration Task-based users Typical users Advanced users
LX Pro: Quad processor
Pentium Pro 200 MHz
768MB RAM
100 60 30
LPr: Dual processor
Pentium II 450 MHz 768MB
RAM
110 65 32
LH 3: Dual processor
Pentium II 450 MHz 768MB
RAM
110 65 32
LXr 8000: Quad Processor
Pentium II Xeon™400
MHz 1GB RAM
200 120 60
Table 2
Resource requirements can vary significantly depending on the application mix
and client usage pattern. Therefore, whenever possible, HP recommends
sizing the server using the targeted application environment.

Hewlett-Packard Company 6
Our tests also show that TSE scales well, evenly sharing the processors
among the client sessions.
The following chart shows the average and maximum processor utilization for
each of the four processors during the 1, 12, 24, 36, 48, 60 and 72 client runs
on the HP NetServer LXr 8000. This chart illustrates how evenly processing is
shared among the processors.
% Processor Utilization
0
10
20
30
40
50
60
70
80
90
100
Number of Client Sessions
Processor 0(Avg) 0.01 0. 05 0. 3 1.66 5. 19 11.14 25.01 37.64
Processor 1(Avg) 0.04 0.08 0.37 2.02 4.77 11.81 25.26 37.66
Processor 2(Avg) 0.01 0.04 0.45 1.68 5.18 11.34 25.52 38.05
Processor 3(Avg) 0.1 0. 03 0.59 1.81 5.13 11.63 25.35 37.79
Processor 0(Max) 0.21 0.78 3.59 6.56 27.99 61.94 100 100
Processor 1(Max) 1.2 1.46 4.35 11. 03 27.34 62.4 100 100
Processor 2(Max) 0.42 0.47 5.47 7.15 28.41 63.85 100 100
Processor 3(Max) 1.09 0.52 5.93 7.54 28.12 63.34 100 100
0 1 12 24 36 48 60 72
Figure 1

Hewlett-Packard Company 7
Memory
As with processors, configuring the server with insufficient memory will
seriously impact performance. Since memory requirements will vary, it is
extremely important to accurately size your server configuration using the
targeted application environment.
For Terminal Server, Microsoft requires 32MB of RAM, plus an additional 4 to
8MB RAM for each user to support running the user's applications on the
server (Microsoft Window NT Server, Terminal Server Administrator's Guide).
Memory utilization during our benchmark test averaged between 5 and 9MB
per user.
Network
As with processor and memory utilization, network utilization will vary
depending on user activity. As a general rule of thumb, Microsoft recommends
2 - 6Kbps per user (Microsoft Terminal Server Capacity Planning, White
Paper).
Network activity during our benchmark test averaged between 2 and 5Kbps per
user.

Hewlett-Packard Company 8
Client Solutions
Microsoft Terminal Server supports a full range of client choices including
Windows-based Terminals (WBTs), Palmtop PCs, NetPCs, and PCs. This
flexibility allows customers to select the client technology that best meets their
needs. HP, as a leading provider of both client and server technology, is
uniquely positioned to offer a full range of client/server solutions for the
Microsoft Terminal Server market.
Our tests have shown that the type of client (i.e., WBT, NetPC, or PC) is not a
significant factor when sizing the server. In order to illustrate this result the
same test script was run on both a PC configuration and a WBT configuration.
12
NetSe rverLHx
10 Mbits
Microsoft Terminal Server
HP NetServer
Microsoft Terminal Server Clients
HP Vectra
HP J31218A
PC Configuration
12
NetServerLHx
10 Mbits
Microsoft Terminal Server
HP NetServer
Microsoft Terminal Server Clients
HP Windows-based Terminals
HP J31218A
WBT Configuration
The next three charts illustrate that server behavior (i.e., performance and
resources utilization) does not vary significantly with a change of client
technology.
The following chart shows the average elapsed time recorded during 1, 2 and 3
client runs using the HP Vectras and HP Windows-based Terminals.
Average Elapsed Time
2000
2100
2200
2300
2400
2500
2600
2700
2800
2900
3000
Number of Client Sessions
PCClients 2440 2435 2431
WBTClients 2444 2438 2434
123
Figure 2

Hewlett-Packard Company 9
The following chart shows the average processor utilization recorded during 1,
2 and 3 client runs using the HP Vectras and HP Windows-based Terminals.
Average Processor Utilization
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Numberof Client Sessions
PCClient s 0.12 0.23 0.34
WBTCl ient s 0.13 0.23 0.35
123
Figure 3
The following chart shows the % Network Utilization recorded during 1, 2 and 3
client runs using HP Vectra and HP Windows-based Terminals.
Network Segment - Average % Network
0
0.5
1
1.5
2
2.5
3
N
umber of Client
PC 0.1
0
.26
0
.37
WBT 0.1
0
.26
0
.37
1
2
3
Figure 4

Hewlett-Packard Company 10
3 Sizing Activity
Methodology
This sizing activity consisted of running the same test script on various HP
NetServer configurations and then comparing the results.
A Visual Test 4.0 script, based on the Microsoft Office 95 product suite, was
used to automate and standardize the interaction of users (i.e., clients) with
Microsoft Windows NT 4.0 Server, Terminal Server Edition using the Microsoft
RDP protocol. (Visual Test is a product of Rational Software,
www.rational.com.)
For each TSE client session, the time to execute the test script (i.e., the
elapsed time) was recorded. In addition to elapsed time, additional system
utilization statistics were gathered using the Microsoft Performance Monitor.
Test Configuration
The test configuration consisted of an HP NetServer connected to 1 - 12 HP
Vectra PCs. PCs, rather than WBTs, were used in the test environment
because the PCs allowed multiple TSE client sessions to be simulated using a
single client system.
To create the network, two sets of six clients were each connected to cascaded
HP J3128A AdvanceStack 10Base-T hubs.
12
NetServerLHx
10 Mbits
Microsoft Terminal Server
HP NetServer
Microsoft Terminal Server Clients
HP Vectra
HP J31218A HP J3128A
Windows NT
(6 workstations) Windows 95 (6 workstations)
Figure 5

Hewlett-Packard Company 11
Microsoft Terminal Server Configurations
Server Softare
• Microsoft Windows NT 4.0 Server, Terminal Server Edition
• Rational Visual Test 4.0
• Microsoft Office 95
Server Hardware
HP NetServer LX Pro
• Quad Intel Pentium Pro 200 MHz processor with 66 MHz bus
• 512KB cache per processor
• 2GB RAM
• Dual, integrated Ultra Wide SCSI controllers - 4.2GB disk drive
• HP DeskDirect 10/100 TX PCI LAN Adapter
HP NetServer LPr
• Dual Intel Pentium II 450 MHz processors with 100 MHz bus
• 512KB cache per processor
• 1GB RAM
• Integrated Ultra2 SCSI controller - 9.1GB disk drive
• HP DeskDirect 10/100 TX PCI LAN Adapter
HP NetServer LH 3
• Dual Intel Pentium II 450 MHz processors with 100 MHz bus
• 512KB cache per processor
• 1GB RAM
• Integrated dual-channel HP NetRAID disk array controller
• Integrated dual-channel Ultra2 SCSI controllers
• HP DeskDirect 10/100 TX PCI LAN Adapter
HP NetServer LXr 8000
• Four Intel Pentium II Xeon 400 MHz processor with 100 MHz bus
• 512KB cache per processor
• 1GB RAM
• Integrated dual Ultra2 SCSI controllers - 4.2GB disk drive
• HP DeskDirect 10/100 TX PCI LAN Adapter

Hewlett-Packard Company 12
Client Configurations
In addition to the Windows-based Terminal configuration, the client
configurations consisted of a mixture of Pentium-based workstations running
either Windows 95 or Windows NT.
HP GT210 Windows-based Terminal
• 16MB RAM
• 10 Base-T Ethernet
• Windows CE 2.10
• Microsoft Remote Desktop Client (RDP)
HP Vectra VL5
• Intel Pentium/133 processor
• 32MB RAM
• HP DeskDirect 10/100 TX PCI LAN Adapter
• Windows NT 4.0
HP Vectra VL5
• Intel Pentium/200 processor
• 32MB RAM
• HP DeskDirect 10/100 TX PCI LAN Adapter
• Windows 95

Hewlett-Packard Company 13
Network
The LAN configuration consisted of two cascaded HP AdvancedStack 10Base-T
hubs. TCP/IP was used as the transport protocol between the TSE clients and
the Microsoft TSE server. The test configuration network was isolated and not
connected to any other network segment.
Under TSE, the client application actually runs on the server. As the application
runs, the server intercepts the application’s user interface (display, keyboard
and mouse) data, and transmits this data between the TSE server and the TSE
client using the RDP protocol.
Microsoft Terminal Server
Application
Presentation
Services Layer
Protocol
Drivers
Transport
Driver
Screen Output
Keyboard/Mouse Screen Display
Presentation
Services Layer
Protocol
Drivers
Transport
Driver
Terminal Server Client
RDP Data Stream
Protocol Packets
Data Transport
12
NetServer LHx
Figure 6

Hewlett-Packard Company 14
Visual Test Scripts
For each client session, a Visual Test 4.0 script was used to automatically
generate keystrokes. Each client session ran the same script.
The script used both Microsoft Word and Microsoft Excel to simulate heavy
keystroke activity, pausing only 0.05 seconds between keystrokes that
generate generic text and 2 seconds between keystrokes that initiate menu
operations.
NOTE The delays used in the test scripts simulate typing at speeds of
approximately 60-70 WPM during the Microsoft Word portion of the
test script.
The Microsoft Word portion of the script creates two 4-page documents. It also
makes various format changes (e.g., font changes and page layout changes).
The Microsoft Excel portion of the script performs numerous calculations (e.g.,
SUM, PRODUCT, and AVERAGE) and generates several charts. Documents
were saved under both Microsoft Word and Microsoft Excel.
NOTE The Microsoft Excel portion of the test has a higher percentage of
menu operations than the Microsoft Word portion of the script. Since
the delay following keystrokes is significantly smaller (i.e., 0.05
seconds) than the delay following menu operations (i.e., 2.0 seconds),
CPU utilization of the Microsoft Word portion of the test is consistently
higher.
The Microsoft Word and Microsoft Excel applications were started at the
beginning of the test script and exited at the end of the script.

Hewlett-Packard Company 15
The following two screenshots were captured during a run of the Visual Test
script.
Figure 7

Hewlett-Packard Company 16
Test Procedures
The server and all clients were rebooted prior to each test run. Once the
server and clients had successfully booted, the required number of TSE client
sessions (i.e., 1-6 client sessions per client workstations) was started on each
client. Then, the Performance Monitor was started as a background task on
the server. Each Visual Test script was then started manually, as quickly as
possible. Once all the clients had completed execution of the script, the
Performance Monitor was stopped and all temporary files were deleted from
the server.
Test Results
This section characterizes the behavior (i.e., elapsed time, processor
utilization, network utilization and memory utilization) of HP NetServers with
various client loads (i.e., 1, 12, 24, 36, 48, 60 and 72).
In order to establish an application baseline (0) value, the script was run at the
server and the Performance Monitor was used to monitor server activity. The
application baseline value is included on some of the charts under the "0"
column.

Hewlett-Packard Company 17
Elapsed Time
An elapsed time was recorded for each client session. As part of the Visual
Test script, the elapsed time was computed and stored in a file on the server’s
disk. Elapsed time was calculated by subtracting the time the script started
from the time the script finished.
The elapsed time includes the time required to load both the Microsoft Word
and Microsoft Excel applications. It does not include the time required to
establish the TSE client session on the server.
The following chart shows the minimum elapsed time recorded when running 0,
1, 12, 24, 36, 48, 60 and 72 client sessions for each of the tested server
configurations.
Minimum Elapsed Time
0
1000
2000
3000
4000
5000
6000
Number of Cl ient Sessi ons
LX Pr o 2396 2353 2354 2401 2543 2899 3213
LPr 2436 2440 2409 2406 2428 2491 2608 2846
LH3 2433 2436 2404 2409 2439 2545 2650 2868
LXr 8000 2433 2437 2411 2399 2399 2398 2445 2454
0 1 12 24 36 48 60 72
Figure 8
The following chart shows the average elapsed time recorded when running 0,
1, 12, 24, 36, 48, 60 and 72 client sessions for each of the tested server
configurations.
Average Elapsed Time
0
1000
2000
3000
4000
5000
6000
Number of Client Sessions
LXPro 2396 2359 2363 2470 2668 3010 3371
LPr 2436 2440 2412 2416 2439 2553 2751 2953
LH3 2433 2436 2418 2421 2445 2540 2750 3011
LXr 8000 2433 2437 2421 2411 2411 2421 2507 2640
0 1 12 24 36 48 60 72
Figure 9

Hewlett-Packard Company 18
The following chart shows the maximum elapsed time recorded when running
0, 1, 12, 24, 36, 48, 60 and 72 client sessions for each of the tested server
configurations.
Maxim um Elapsed Tim e
0
1000
2000
3000
4000
5000
6000
Number of Client Sessions
LXPro 2396 2364 2376 2502 2746 3123 3517
LPr 2436 2440 2420 2423 2449 2636 2892 3022
LH3 2433 2436 2422 2429 2453 2619 2841 3128
LXr 8000 2433 2437 2432 2421 2424 2491 2624 2920
0 1 12 24 36 48 60 72
Figure 10
Process Utilization
Processor utilization was monitored and recorded using the Microsoft
Performance Monitor.
Processor - % Processor Time
The % Processor Time counter describes the percent of time a particular
processor is busy.
System - % Total Processor Time
The % Total Processor Time counter describes the percent of time all
processors are busy. In a multiprocessor computer, this counter is equal to
the sum of % Processor Time for each processor divided by the number of
processors. If the server workload is shared equally among all processors,
System: % Total Processor Time is a good measure of processor activity.
System - Process Queue Length
Processor Queue Length is the instantaneous length of the processor queue
in units of threads. All processors use a single queue in which threads wait
for processor cycles. This length does not include the threads that are
currently executing. This is an instantaneous count, not an average over the
time interval.

Hewlett-Packard Company 19
The following Performance Monitor charts show % Processor Time while
performing the 12 client session tests on the HP NetServer LH 3.
Figure 11 - 12 Client Sessions (LH 3)
The following Performance Monitor charts show % Processor Time while
performing the 36 client session tests on the HP NetServer LH 3.
Figure 12 - 36 Client Sessions (LH 3)

Hewlett-Packard Company 20
The following Performance Monitor chart show % Processor Time while
performing the 72 client session tests on the HP NetServer LH 3.
Figure 13 - 72 Client Sessions (LH 3)
NOTE As CPU utilization increases, with the increased number of users, the
time required to complete the Microsoft Word portion of the script
increases. The average time required to complete the Microsoft Word
portion of the script on the LH 3 was 897 seconds for the 12 client run,
919 seconds for the 36 client run and 1361 seconds for the 72 client
run. The average time required to complete the Microsoft Excel
portion of the script was 1521 seconds for the 12 client run, 1534
seconds for the 36 client run and 1650 seconds for the 72 client run.
As discussed previously, this is because the Microsoft Word portion of
the script actually places a heavier load on the system and saturates
the processors before the Microsoft Excel portion of the script.