Test Lab - Compaq 108164-003 - ProLiant - 800 White Paper

W P
HITE APER
(cont.)
. .
. .
. .
ProLiant 7000 PS Bottlenecks
. .
. .
Analysis of PS bottlenecks on the ProLiant 7000 shows:
. .
. .
. .
. .
Table 16: ProLiant 7000 Analysis of PS Bottlenecks
. .
. .
Resource Bottleneck
. .
. .
. . CPU No
. .
. .
. .
Network No
. .
. .
. .
. .
Memory No
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
. .
Operating No
. .
System
. .
. .
. .
Web Server Possible
. .
. .
Software (IIS)
. .
. .
PS Software Possible
. .
. .
. .
. .
Disk Yes
. .
Subsystem
. .
. .
. .
. .
. .
. .
. .
. .

Test Lab

. .
. .
. .
The test lab configuration used for the ProLiant 2500 testing is shown in Figure 5. The number of
. .
client systems needed was determined by adding one at a time until the ProLiant 2500's CPU was
. .
. .
driven to 100% utilization. Table 17 shows the actual number of client systems used and the
. .
configurations for each test. There was always one network for the ProLiant 2500 tests.
. .
. .
. .
. .
Table17: ProLiant 2500 Test Lab Configuration
. .
. . Test
. .
. .
. .
ILS Connect
. .
ILS Add
. .
. .
ILS Direct Lookup
. .
All PS Tests
. .
. .
. .
. .
. .
. .
18
Comments
The CPU utilization decreases as the number of users in the
UPD increases.
The total network traffic is less than 15% of the available
bandwidth.
There is over 200 MB of memory available. The "Pages/sec"
measurement shows a high page fault rate with larger UPDs.
The cause is most likely that PS is using memory-mapped
files. The Physical Disk perfmon measurements support this
conclusion because the disks are busy four times as much
with reads than they are with writes.
The web server may be a bottleneck as the number of ASP-
enabled web pages requested increases.
PS may be a bottleneck as the number of concurrent users
increases.
See the analysis for the PS Software above
# of Client Systems # of Simulated Users and Worker Threads per Client
System
5
2
4
3
50
20
50
11