HP ML570 - ProLiant - G2 Quick start guide

Novell NetWare 6 performance tuning guidelines for ProLiant
servers
table of contents
december 2002
hp industry
standard servers
integration note
TC021201IN
abstract 3
introduction 3
understanding the server deployment environment 4
why use industry standard benchmark for performance analysis 4
server subsystem components and selection guidelines 5
disk subsystem 5
SCSI specifications 5
monitoring the disk subsystem in NetWare 6 8
tips and tricks 9
network subsystem 12
monitoring the network subsystem in NetWare 6 14
tips and tricks 15
memory subsystem 15
monitoring the memory subsystem in NetWare 6 16
tips and tricks 16
system processor 17
monitoring the system processor in NetWare 6 18
tips and tricks 19
introducing NetWare 6 19
setup / installation checklist 20
tuning guidelines 20
NetWare performance monitoring tools 21
Monitor 21
tips and tricks 22
performance parameters 22
NetWare Remote Manager (Version 1.7.3) 25
NRM tool utilities 26
Vtune 30
introducing NetBench 31
introducing WebBench 32
NetBench test results 33
processor scalability performance 33
performance effects of client side file caching 34
performance effects of TFS versus NSS 35
performance effects of tuning 36
performance effects of hyper-threading 36
WebBench test results 40
1

Novell NetWare 6 performance tuning guidelines for ProLiant servers
processor scalability performance 40
default installation versus optimized parameters 41
number of threads per client 42
conclusion 43
appendix a: server features 44
appendix b: NetWare 6 configuration files 45
appendix c: NetBench test methodology 46
appendix d: WebBench test methodology 47
appendix e: server configuration details 48
appendix f: test bed details 49
references 50
for more information 50
feedback 51
2

Novell NetWare 6 performance tuning guidelines for ProLiant servers
abstract IT (information technology) professionals and administrators are constantly under intense
pressure to cut cost, protect current investments, and meet immediate performance
demands without service interruption. Achieving these requirements within tight budgetary
constraints is not an easy task. In order to optimize your investment, it is important the
server operates at the highest performance level possible.
To achieve optimum performance, a server must contain key subsystem components in
addition to being configured and tuned correctly. This document provides information on
the important subsystem components to consider and suggestions on how to configure and
tune them.
Given that the operating system and hardware components contribute to overall server
performance, the focus of this document will be on the configuration and tuning of the
following key components:
• network operating system (NOS) – NetWare 6
• hardware components – disk, network, memory, processor, bus interface
The NetBench and WebBench test results sections of this document also provide sample
performance results to reinforce some of the recommendations and suggestions provided in
this document.
introduction Determining the correct hardware / software mix for the user, takes in depth knowledge of
the environment in which the server will be deployed. This key element is vital in order to
properly tune and configure the server to maximize performance and eliminate
bottlenecks. Knowledge of the user environment includes the following aspects:
• user needs – what the user requires in terms of security, storage space, etc.
• user workload characterization – types of application that will be running
• user performance requirements – level of response times / throughput expectation
• budget – the amount of money allocated for the project
A thorough understanding of the user environment aids the IT professional or administrator
in selecting the correct hardware and software mix that meets the requirements without
exceeding the budget.
In this document the Ziff-Davis industry standard benchmark is used as the workload
generator. The NetBench and WebBench test suites are used to simulate “real-world” file
I/O requests and web application workloads respectively. The operating system platform
chosen is Novell NetWare 6 and the hardware selected is the ProLiant ML570 Generation
2 (G2) server.
Although the server used in this document is the ProLiant ML570 G2, the knowledge
gained as a result of this study can easily be applied to any HP servers deployed in similar
environments.
3

Novell NetWare 6 performance tuning guidelines for ProLiant servers
understanding
the server
deployment
environment
Generally speaking, a server can typically be deployed in any environment the user
chooses. However, server performance varies depending on configuration, operating
environment, and workload.
Typically, in a given platform, the server subsystems likely to be exercised are the
processor, memory, network, and the disk. However, the extents to which any of these are
exercised depend largely on the workload characteristics and the operating system
environment in which the server is deployed. For example, under a database workload,
the processor, memory, and disk subsystem of the server come under intense pressure;
whereas in file and print service environments, the network, memory, and disk are
intensely exercised. Similarly, under web applications, the network, memory, and
processor come under intense pressure. Likewise, under Exchange/Messaging workload,
the memory, processor, and disk are greatly stressed. And lastly, under an application
server workload, the memory, processor, and disk are under intense pressure.
To achieve optimum server performance, it is important to understand the particular
environment the server will operate in and the impact the server subsystem components
selected might have on the server’s overall performance. Having this knowledge enables
you to fine-tune the server appropriately; thereby isolating and eliminating performance
bottlenecks.
This document provides a general overview of typical server subsystem components and
the tuning guidelines for deployment in a file I/O and web application environments. The
Ziff-Davis NetBench and WebBench benchmarks running under NetWare 6 operating
system are used as the basis for the analysis.
why use industry
standard benchmark
for performance
analysis
Ideally, the best benchmark should be the same exact application the user or customer will
be running on his/her platform. However, this is usually not possible in most situations.
Therefore, most IT professionals use an industry standard benchmark application that best
simulates their unique environment in order to predict how well a server will perform when
deployed.
In general, benchmarks can be categorized into two main groups:
• trace driven
• execution driven
The trace driven benchmark focuses primarily on real-world performance using
capture/playback “traces” of real-world applications. The Ziff-Davis benchmark test suites
fall in this category. This benchmark is widespread in the industry and is believed to mimic
real-world applications.
The execution driven benchmark is synthetic and tends to focus on certain aspects of the
server subsystem (e.g., processor, memory, etc.) that does not usually correlate to real-
world usage experience. The SpecCPU2000, for instance, is a good example of a
synthetic benchmark.
The object of this discussion is not to preclude or endorse one category over the other but
to highlight the differences so the user can make an informed decision as to which method
accomplishes set goals.
4

Novell NetWare 6 performance tuning guidelines for ProLiant servers
As stated earlier, the Ziff-Davis test suites were chosen. The decision to use the Ziff-Davis
Benchmark is because it is designed to mimic real-world user applications. Original
equipment manufacturers (OEMs) and personal computer (PC) magazine use the Ziff-Davis
test suites in their research publications. To obtain the latest copy, go to www.zdnet.com.
server subsystem
components and
selection guidelines
The IT professional or the system administrator is responsible for selecting which subsystem
components provide sufficient bandwidth for optimal overall server performance.
Once the server is selected, the appropriate NOS is chosen to leverage the server
hardware components. For instance, if a multiprocessor (MP) server is selected, then a
corresponding network operating system and applications that are MP-aware should be
used.
The following sections highlight examples of critical server subsystem components and how
the selection you make can dramatically affect server performance.
disk subsystem The disk subsystem consists of the disk and the controller. One of the functions of the disk
subsystem is to keep the memory or cache filled with useful data. Insufficient memory leads
to a low cache hit rate and slower user access time. When this occurs, performance
decreases. It is a known fact that disk response time is significantly slower than memory
response time.
The performance of the disk subsystem can therefore be affected by the components that
constitute the subsystem. There are several elements that makeup the disk subsystem (i.e.,
SCSI (small computer system interface), disk, controller, drivers, bus interface, etc.). The
following components are crucial during the selection process:
• SCSI
• disk
• controller
• SCSI technology
SCSI technology is a widely used standard for transferring data between the disk drive
and the SCSI controller. Many types of SCSI technology are currently available and the
one you chose can impact overall system performance. It is therefore recommended to
select the type with the highest speed your disk drive and controller can support. Table 1
lists the different types of SCSI standards currently available.
SCSI specifications table 1. SCSI standard specifications
SCSI standard bus speed
(MHz)
50-pin narrow
(8-bit)
68-pin wide
(16-bit)
maximum cable
length
SCSI 5 5 MB / s 6 meters
SCSI-2 10 10 MB / s 20 MB / s 3 meters
SCSI ultra 20 20 MB / s 40 MB / s 1.5 meters
SCSI ultra2 40 40 MB / s 80 MB / s 12 meters (LVD)
SCSI ultra3 40 160 MB / s 12 meters (LVD)
SCSI ultra4 80 320 MB / s 12 meters (LVD)
5

Novell NetWare 6 performance tuning guidelines for ProLiant servers
There have been significant improvements in drive technology over the past decade. For
instance, disk drive media speeds continue to increase. HP offers a wide range of drive
spindle speeds in revolutions per minute (rpm). For instance, there are 10,000 and
15,000 rpm disk drives. The 15,000 rpm models are available in 18.2 GB and 36.4 GB
with an average seek time of 3.8 ms, while the 10,000 rpm models are available in
18.2 GB, 36.4 GB, and 72.8 GB with an average seek time of 5.5 ms. It is therefore,
advisable to choose the appropriate disk type with the highest speed your SCSI interface
and controller can support.
Similarly, as shown in table 2, HP has both Smart Array and Non Array controllers; all
with different performance and capacity. The non-arrays such as the Integrated Drive
Electronics (IDE) controllers and the low-end array controllers are typically embedded on
the system board.
table 2. Smart Array and Non Array controllers
Smart Array
controllers
channels cache
(read/write)
max drive
supported
SCSI protocol
supported
compatibility
table
Smart Array
5312
2 128 28 Ultra3
Ultra2
Smart Array
5304
4 256 128 56 Ultra3
Ultra2
Wide Ultra
support for
ADG
Smart Array
5302
2 128 64
32*
28 56** Ultra3
Ultra2
Wide Ultra
optional
support for
ADG
Smart Array
4250ES
3 64 21 Ultra2
Wide Ultra
Smart Array
532
2 32* 20 Ultra3
Ultra2
Smart Array
431
1 16 14** Ultra3
Ultra2
Wide Ultra
Smart Array
5i plus
2 64* 20 Ultra3
Ultra2
Smart Array
5i
2 32* 20 Ultra3
Ultra2
Integrated
Smart Array
1 16* *** Ultra2
Wide Ultra
*- used for code, transfer buffers, and non-battery backed cache
**- based on use of HP storageWorks 14 HDD enclosure
***- depends on server
6

Novell NetWare 6 performance tuning guidelines for ProLiant servers
For more information regarding HP Array controllers, visit:
h18000.www1.hp.com/products/servers/proliantstorage/arraycontrollers/index.html.
The Redundant Array of Inexpensive Disks (RAID) controllers can be configured in any of
the following supported RAID levels depending on the performance level and data
protection desired.
• RAID 0 – No data protection. Simply stripes data across multiple disk drives without
any data protection. This configuration provides high level of performance at low
cost.
• RAID 1 – Disk mirroring. Requires twice the amount of disk drives in order to
duplicate user data for data protection. This configuration provides similar level of
performance as RAID 0 in read operations. The cost is, however, higher because of
the duplication of disks.
• RAID 5 – Distributed data guarding. In this configuration, the parity data is
distributed across all drives. This provides protection against the failure of any one
drive in the array set. It provides improved performance at a minimum cost.
• RAID Advanced Data Guarding (ADG) – In this configuration, two sets of parity data
is distributed across all drives. This provides the protection against the failure of any
two disk drives in the array set. This provides high fault tolerance at a minimum cost
of implementation.
The disclosure of the server configurations for this document can be found in appendix e.
The disk drive, SCSI channel, and disk controller all have to communicate with each other
during data transmission. Realize that these devices transfer data at different rates. It is
therefore recommended to balance out the I/O devices on the buses at the right speed for
optimum performance benefits.
The personal computer interconnect (PCI) / PCI-eXtension (PCI-X) bus is a 32-/64-bit
industry standard interface for high-speed data transfers between peripheral components.
For example, the standard PCI clock rate can be 33 or 66 MHz. On the other hand PCI-X,
which is an extension of the PCI, operates at up to 133 MHz. PCI-X devices are backward
compatible and adjust to the bus speed of the vendor implementation. HP servers support
several implementations. Table 3 provides some examples of theoretical PCI / PCI-X
transfer rates.
table 3. theoretical maximum achievable PCI / PCI-X transfer rates
PCI bus type
clock transfer
rate PCI (32bit) PCI (64bit) PCI-X (32bit) PCI-X (64bit)
33 MHz bus 132 MB / s 264 MB / s 132 MB / s 264 MB / s
66 MHz bus 264 MB / s 528 MB / s 264 MB / s 528 MB / s
100 MHz bus 400 MB / s 800 MB / s
133 MHz bus 532 MB / s 1064 MB / s
7

Novell NetWare 6 performance tuning guidelines for ProLiant servers
For example, the theoretical maximum achievable throughput on a 66 MHz PCI when a
32-bit bus master bursting with a 32-bit target would be:
4 bytes per data phase x 66 million data phases per second = 264 MB / second
monitoring the disk
subsystem in
NetWare 6
Utility tools such as Monitor, NetWare Remote Manager (NRM), and SET parameters can
be used to gauge disk subsystem performance in NetWare 6. These utilities provide clues
as to whether there is a performance bottleneck in the disk subsystem. The following are
parameters to monitor:
• dirty cache buffers – Number of cache buffers containing updated data that has not
been written to the disk. The operating system writes the data to the disk either as
soon as the cache buffer is filled or when the dirty cache delay time elapses (default
3.3 seconds).
• current disk requests – Number of pending disk I/O requests that are queued for
service.
• least recently used (LRU) settings – measures the length of time it takes for a cache
buffer at the head of the list to make its way down to the tail, where it becomes the
LRU buffer.
• Novell Storage Services (NSS) cache buffers – Check the cache buffers used by the
NSS. To view the statistics shown in table 4, type the following at the console prompt:
NSS / Cachestats
table 4. buffer cache statistics
statistics cache
num cache buffer 512
num hash buckets 524288
min OS free cache buffers 256
num cache pages allocated 233106
cache hit percentage 76
cache hit 69803
cache miss 21624
percent of buckets used 0%
max entries in a bucket 1
total entries 1286
Check the values for the Cache hit percentage and the Percent of buckets used on the
displayed result.
• Cache hit percentage provides a percentage of how the cache is performing. A
percentage value of > 75 is desirable.
• Percent of buckets used provides a percentage of the total cache currently in use.
8

Novell NetWare 6 performance tuning guidelines for ProLiant servers
tips and tricks Use the following tips to improve the performance of your disk subsystems:
• If the number of dirty buffers remains constant and the number of current disk
requests remains high, the disk subsystem might be a bottleneck. Consider installing a
faster disk drive and a controller.
• If the number of dirty buffers is frequently above 50% of total cache buffers, add
more memory for cache.
• Monitor the swap activities (listed in table 5) taking place. To view the swap file,
type the following console command:
swap
table 5. virtual memory swap file (sizes in millions of bytes)
volume name: SYS
swap file volume
size: 22 size: 4116
used: 0 free space: 2889
min: 2 min free: 5
max: 4116
Summary
total swap file total volume
size: 22 size: 4116
used: 0 free space: 2889
min: 2 min free: 5
max: 4116
• If necessary, create a swap file (one per volume) by typing the following console
command:
SWAP ADD SYS1 MIN=4, MAX=8, MIN FREE=3000
MIN = minimum swap file size (default = 2)
MAX = maximum swap file (default = free volume space)
MIN FREE = minimum free space to be preserved on a volume outside the swap file
(default = 5).
• To remove a swap partition, type the following console command:
SWAP DEL SYS1
The Swap file is a storage location for data moved to the disk by the virtual memory
manager. To monitor Swap file activities using the NRM tool, follow these steps:
1. Open the NRM and select View Memory Config listed under the Manage Server
category.
Note: For detailed instructions on how to invoke the NRM, refer to the NetWare
Remote Manager section of this document.
9

Novell NetWare 6 performance tuning guidelines for ProLiant servers
10
2. Select Swap File Size in the System Memory Information window to view the swap
file usage information.
3. Scroll down with the right arrow to display the Total Disk Usage in Pie chart as
shown below.

Novell NetWare 6 performance tuning guidelines for ProLiant servers
To monitor disk activities using the NRM tool, follow these steps:
1. Open the NRM, and then select Health Monitor listed under the Diagnose Server
category to display the following screen.
Note: For detailed instructions on how to invoke the NRM, refer to the NetWare
Remote Manager section of this document.
2. Check Disk Throughput (as shown in the figure above) and click Apply.
3. Click on Disk Throughput for a graphical display of the Disk Throughput information
as shown in the next figure.
11

Novell NetWare 6 performance tuning guidelines for ProLiant servers
network subsystem The network subsystem is essentially the server’s interface to other computers (i.e., LAN
clients). One of its primary functions is to move data between the network and system
memory as fast as possible. The network subsystem is crucial in heavy file I/O and print
applications, where the majority of the request sizes are of small record sizes.
Therefore the performance of the network subsystem can be impacted by the selection and
configuration of the components that makeup the subsystem. The main network subsystem
components are:
• topology – Ethernet, Token Ring, Fiber Distributed Data Interface (FDDI),
Asynchronous Transfer Mode (ATM), etc.
• network interface card (NIC) – 32 bit vs. 64 bit, PCI NIC vs. PCI-X NIC
• network design – switches vs. hubs
The Ethernet topology is a widely used standard topology in local area networks (LANs)
compared to Token Ring, FDDI or ATM. The Ethernet technology has evolved rather rapidly
over the years. It is less expensive to implement and it is backward compatible. Token Ring
has become less popular due bandwidth limitation and cost. FDDI is a high speed
interface, however, it is not as widely used in LANs because it is complicated and
expensive to implement. The ATM, on the other hand, is mainly used as backbones and for
wide area networks (WAN).
Ethernet NICs are available in a variety of types. There are PCI / PCI-X models with
support for 10/100/1000 Mbps speed. The 802.3ae standard for 10-Gigabit Ethernet
equipment was adopted by the Institute of Electrical and Electronics Engineers, Inc. (IEEE)
in June 2002. For more information on 802.3ae, visit
grouper.ieee.org/groups/802/3/ae/.
12

Novell NetWare 6 performance tuning guidelines for ProLiant servers
HP has many different NICs with different speeds, bandwidth, and chipsets as shown in
table 6. Thus, they do have different features and varying degrees of performance and
cost associated with them. For detailed information and product ordering visit:
ftp.compaq.com/pub/products/servers/networking/model-compare.pdf.
table 6. hp available NICs
model
number
description transfer rate bandwidth / bus
MHz
chipset
NC3123 single port PCI
10/100 (WOL)
10, 100 Mbps 32 bit / 33 MHz Intel 82559
NC3133 single 100FX
upgrade module
100 Mbps 32 bit / 33 MHz Intel 82558
NC3134 dual port 10/100
base adapter
10, 100 Mbps 64 bit / 66 MHz Intel 82559
NC3135 dual port 10/100
upgrade module
10, 100 Mbps 32 bit / 33 MHz Intel 82559
NC6132 single port 1000 SX
upgrade module
(fiber)
1000 Mbps 64 bit / 33 MHz Intel 82542
NC6133 single port 1000 LX
upgrade module
(fiber)
1000 Mbps 64 bit / 33 MHz Intel 82542
NC6136 single port 1000 SX
(fiber)
1000 Mbps 64 bit / 33 MHz Intel 82543GC
NC7770 PCI-X single port
copper Gigabit
(copper)
10, 100, 1000
Mbps
64 bit / 133
MHz
Broadcom
5701(h)
NC7131 single port
10/100/1000-T
(copper)
10, 100, 1000
Mbps
64 bit / 66 MHz Intel 82543GC
NC7132 10/100/1000-T
upgrade module for
NC3134 (copper)
10, 100, 1000
Mbps
64 bit / 33 MHz Intel 82543GC
Likewise, HP has a variety of cost effective Ethernet switches with performance ranging
from 10 Mbps to 10 / 100 Mbps to 10 / 100 / Gigabit that are either managed or
unmanaged. Therefore, it is recommended to match-up the switches with the right interface
devices for maximum performance benefits. Also, be sure to extend the network traffic
across multiple buses on separate LAN segments to maximize throughputs and minimize
collisions.
For more information on HP access/distribution switches, visit:
www.hp.com/rnd/products/switches/.
13

Novell NetWare 6 performance tuning guidelines for ProLiant servers
monitoring the
network subsystem in
NetWare 6
In NetWare 6, use the monitor, NetWare Remote Manager (NRM), and the SET NetWare
utility tools to gauge the status of the network subsystem performance. These utilities
provide clues as to whether there is a network subsystem bottleneck in your system. Some
of the parameters to watch closely are:
• packet receive buffers – number of buffers that are available to the file system for
holding client requests until they can be processed.
• maximum packet receive buffers – specifies the maximum number of packet buffers
the operating system can allocate.
• maximum physical receive packet size –specifies the maximum packet size that can
be transmitted on the network.
• NetWare Core Protocol (NCP) packet signature option – controls the NCP packet
signature level on the server.
To monitor LAN Traffic using the NRM tool, follow these steps:
1. Invoke the NRM, and then select Health Monitor listed under the Diagnose Server
category to display the following screen.
Note: For detailed instructions on how to invoke the NRM, refer to the NetWare
Remote Manager section of this document.
2. Check LAN Traffic (as shown in the figure above) and click Apply.
3. Click on LAN Traffic for a graphical display of the Packets Per Second as shown in
the next figure.
14

Novell NetWare 6 performance tuning guidelines for ProLiant servers
tips and tricks Use the following tips to improve the disk subsystem performance:
• Use fast 64-bit PCI/PCI-X Direct Memory Access (DMA) NIC devices with gigabit
adapters in the server, if available.
• Increase the maximum packet receive buffers (in increments of 10), until you have
one packet receive buffer per workstation.
• Set the maximum physical receive packet size to the correct size for your topology.
For Token Ring and Ethernet topologies, 4202 is adequate.
• Set the NCP packet signature option to 0 to reduce network traffic congestion and
server processor consumption.
• Use network monitoring device to observe network segment utilization. Add another
segment, if needed.
memory subsystem The amount of available memory for network buffers and disk I/O caching can profoundly
affect the performance of a server. Typically the amount of memory in a server should be
proportional to the number of users. That is, the higher the number of users, the larger the
amount of memory required. Low memory, for instance, could lead to “disk thrashing”
which translates to slower response times and low throughputs.
Memory is available in many different types. The differences in the technology and speed
affect memory access time and the overall server performance. HP servers support several
memory technologies. The goal of this section is to create an awareness of the existing
types of memory technologies that are currently available. Table 7 provides information on
different memory technologies and the potential performance implications.
15

Novell NetWare 6 performance tuning guidelines for ProLiant servers
table 7. types of memory technology
type description
extended data out (EDO) This memory technology was first introduced early in
the year 1994 as an improvement over Fast Page
Mode (FPM) memory technology. The clock rate was
40 MHz with a max bandwidth of 320 MB/s.
Synchronous Dynamic
Random Access Memory
(SDRAM)
This memory technology was first introduced in the
early 1996 at 66 MHz. The 100 MHz version was
introduced in the year1998 and in the year 1999,
133 MHz version was introduced. The unbuffered
DIMMs though slightly faster than the register
DIMMs has limitation in the number of DIMMs that
can be interconnected on a bus due to electrical
loading. The registered DIMMs on the other hand
have lighter electrical load requirements and thus
more DIMMs can be interconnected on a bus.
double data rate (DDR) Introduced in the year 2000 as DDR SDRAM (100
MHz / 133 MHz). In theory DDR is capable of
transferring data on both the rising and falling
edges of the clock.
monitoring the
memory subsystem in
NetWare 6
In NetWare 6, use the NetWare Monitor and/or the all-inclusive NRM utility to check the
status of the memory parameters. These utilities provide clues as to whether there is a
memory bottleneck in your system. The following parameters apply to only the traditional
file system. For NSS values, refer to the NSS section in this document.
• long term cache hit – cumulative percentage of the system memory since last started.
• LRU setting time – measures the length of time it takes for a cache buffer at the head
of the list to make its way down to the tail, where it becomes the LRU buffer.
• SWAP activities – monitors the rate at which data is being moved from memory to
disk.
tips and tricks Use the following tips to improve the disk subsystem performance:
• Use the NRM or the NetWare monitor utility to regularly monitor total memory
available and cache statistics.
• Set the long term cache hit counter range to be within 90% to 100%.
• Set the LRU setting time to greater or equal to15 minutes.
• Use the Swap utility to monitor virtual memory performance. If the overall system
memory is running low, swapping does occur more often.
• Add more memory, if swapping or disk thrashing is taking place due to insufficient
system memory.
16

Novell NetWare 6 performance tuning guidelines for ProLiant servers
There are several tools available under NetWare for monitoring memory performance and
the general health of the server. One such tool is the NetWare Monitor, which can be
started by typing “monitor” on the console or through the NRM.
To monitor memory activities using the NRM tool, follow these steps:
1. Invoke the NRM and select View Memory Config listed under the Manage Server
category.
Note: For detailed instructions on how to invoke the NRM, refer to the NetWare
Remote Manager section of this document.
2. Scroll down using the right arrow to display the pie chart of current memory usage
(as shown in the next figure).
system processor In Intel architecture, the processor is one of the most important hardware subsystem
components. The processor is the brain of the system and thus it is involved in most of the
transaction activities that occur within the server. For instance, all instructions generated by
LAN attached clients must first be interpreted and processed by the processor. Thus for a
server configured as a database or application server, the processor becomes critical and
therefore more likely to be the source of bottleneck. Similarly in a file server environment,
the processor is the heart of the file server. However, a file server is more reliant on the
network, memory, and the disk subsystems for throughput capacity.
Intel processors are available in different architectural models and packages. They range
from the 32-bit Pentium II, III, 4, to 64-bit Itanium 2 with front side bus (FSB) speeds of up
to 533 MHz that is 128-bit wide and level 3 cache sizes of up to 3 MB. The cache can be
accessed at full or half the speed of the processor in some cases. Complete details of the
17

Novell NetWare 6 performance tuning guidelines for ProLiant servers
In-depth information of Intel processor architecture is beyond the scope of this document.
Intel has recently introduced a relatively new technology called hyper-threading on their
Xeon based family of processors. This technology exploits a program’s instruction and
thread-level parallelism in order to maximize overall system throughput. The ProLiant
ML570 G2 server supports this technology. For more information on Intel’s hyper-threading
technology, visit: www.intel.com/technology/hyperthread/.
HP servers typically ship with one or more chipsets from several vendors such as Intel,
ServerWorks, etc. The goal of this section is to create a performance awareness to guide
the IT professional or the administrator during the decision / selection process when
buying a server.
monitoring the
system processor in
NetWare 6
In NetWare 6, use the Monitor, NetWare Remote Manager (NRM), and the SET
Parameter utility tools to gauge the status of the processor subsystem performance. These
utilities provide clues as to whether there is a processor subsystem bottleneck in your
system.
You can use the NRM tool to monitor the trend statistics of any subsystem element that is of
interest. For example, to use the NRM to monitor the trend statistics of the processor, follow
these steps:
1. Invoke the NRM, and then select View Statistics listed under the Manage Server
category to display the following screen.
Note: For detailed instructions on how to invoke the NRM, refer to the NetWare
Remote Manager section of this document.
18

Novell NetWare 6 performance tuning guidelines for ProLiant servers
2. Check the time sampling interval of the trend update, and then click on Draw
Selected Graphs for a graphical display of the selected element as shown in the
next figure.
tips and tricks Use the following tips to improve the disk subsystem performance. The processor utilization
parameters to check are:
• processor utilization – a server’s total processing capacity used during the last
second (listed as a percentage). Check the processor utilization regularly. If the
utilization is
> 90% over a sustained period of time, it probably indicates a processor bottleneck.
• watch for errant NetWare Loadable Modules (NLMs).
• make sure the current service processes counter is adequate.
• use Direct Access Memory (DMA) devices.
• upgrade or add another processor, if the processor is the bottleneck.
There are several tools available under NetWare for monitoring processor subsystem
performance and the general health of the server. Refer to the NetWare Performance
Monitoring Tools section of this document for more examples.
introducing
NetWare 6
A
ccordin
g
to Novell, the current version of NetWare 6 is a multithreaded, multitaskin
g
,
MP-enabled operating system. A thread is a stream of control that executes instructions
independently. This stream of instructions consists of tasks in the thread. A multithreaded
program is where two or more threads can execute concurrently. Typically, a
multithreaded program will not automatically run on more than one processor unless
specified by the code developer. A multithreaded operating system with multitasking
capability on the other hand can execute threads from different multithreaded programs
19

Novell NetWare 6 performance tuning guidelines for ProLiant servers
concurrently on a single processor by the means of a round robin technique. An MP-
enabled program has the ability of executing one thread on several processors in a system
at exactly the same time. In order to enhance scalability, NetWare 6 uses per-processor
run queues technique for efficiency and to improve server performance. This method
“affinitizes” threads to processors by allowing threads to run on the same processor they
ran previously. That way, the likelihood of having useful data in the high-speed cache for
immediate access by the threads is improved. For further details, refer to
www.novell.com/info/collateral/docs/4621199.03/4621199.pdf.
For this document, Novell’s NetWare 6 with Support Pack 2 (SP2) was used. According to
Novell, NetWare 6 is optimized out-of-the-box and is also capable of tuning itself
dynamically without user intervention. The results of comparing an out-of-the-box version to
one with adjusted values are provided later in this document. Furthermore, an explanation
of the parameters and the values that were adjusted are also presented. Examples of some
useful tools to monitor and make changes to the parameter values are also provided.
setup / installation
checklist
The following information is a guideline to assist you after you reached the decision that
NetWare is the right operating system for your platform. This list is by no means an
exhaustive one.
• know the environment in which the server is being deployed.
• know what types of applications are going to be run (profile if possible).
• know the number of users and their potential usage requirements.
• identify the disk (RAID), memory, NIC, and processor requirements.
• determine if your hardware/software is Novell “YES” certified. Refer to
www.developer.novell.com for tested and approved listings.
• use the latest hardware and software (device drivers) from all vendors you plan to
use. Visit their website to check for their current product offerings.
• apply the latest support pack and updates.
tuning guidelines According to Novell, NetWare 6 is optimized out-of-the-box by default. However,
depending on the application platform and/or customer needs, you can adjust many
parameter values in order to optimize the performance of your server.
First collect a baseline measurement before making any adjustments to the values in the
parameter lists. That way, you will be able to evaluate whether the changes you made had
any effects on the performance objectives you’re trying to achieve. Some suggestions are
as follows:
• define your performance objectives or requirements.
• record your baseline measurements.
• use the NetWare monitoring tools to regularly observe how the server is performing.
That way, you will be able to spot any imbalance before it becomes a server
bottleneck.
• collect and analyze your measurement results in order to pin point a bottleneck.
Once a bottleneck has been identified, fix it.
• only make one change at a time. Run your benchmark, analyze and record the
results and compare it with the baseline measurement.
• repeat the above exercise until the desired performance levels or objectives are
achieved.
20