Compaq ProLiant 1000 Performance Analysis and Tuning of Raptor's Eagle NT 3.06 - Page 17

MaxRecieve Buffers for NetFlx-3 Cards, Processor

Page 17 highlights

278A/0497 WHITE PAPER (cont.) ... not a big increase in performance since FTP transfers are still doing DNS Lookups. On HTTP only transfers, the TPM and the percent TPM difference increase from the base system is higher because the HTTP daemon supports the switch for no DNS Lookups and FTP currently does not. Please refer to section Test Results with HTTP Only for HTTP only test results. Turning off HTTP caching of the rules, run #6, degrades performance overall by 1% from the base system with a peak low of 4%. The overall average failure rate for both tests is less than 1% from the base system. The performance degradation is expected since caching allows the rule base only to be checked once a minute as opposed to every connection. Allowing HTTP caching of the rules especially improves performance if the rule base is large. MaxRecieve Buffers for NetFlx-3 Cards Setting the MaxRecieve buffers for the NetFlx-3 NIC cards to 500 represented by Run #7 and Run #8. Run #7includes the MaxRecieve Buffers change and 2 PCI Netflx-3 cards. This results in an overall 16% increase from the base system. Run #8 displayed negative differences of 5% from the base system for 1 to 24 virtual clients and increased as the load increased to an average of 5.6% above the base for 32 to 72 virtual clients. The overall average for all percent differences for 1 to 72 virtual clients was 1.6%. Failures decreased from the base system to overall 0.02% for all virtual client runs. Processor In Run #9, a Pentium Pro 200 MHZ-512 cache processor was added to the base system. Review Graph 4 below for results. 1100 1000 900 800 700 600 500 400 300 200 100 0 Run #9 in TPM 1 12 24 32 36 48 56 72 Virtual Clients Base Run 2- PP200, 512c Graph 4: Base Run with 2 Processors Adding the second processor clearly shows performance improvements to the TPM from the base system. The overall average percent difference increase from the base system for all virtual clients is 15%. As the load increased, the TPM is increased bringing the positive percent difference to over 36% from the base system with 72 virtual clients. Failures increased slightly to 2% overall for all virtual clients. The increase in TPM is due to the use of IO Completion ports and threads in the HTTPD of Raptors Eagle NT 3.06 firewall software. 17

  • 1
  • 2
  • 3
  • 4
  • 5
  • 6
  • 7
  • 8
  • 9
  • 10
  • 11
  • 12
  • 13
  • 14
  • 15
  • 16
  • 17
  • 18
  • 19
  • 20
  • 21
  • 22
  • 23
  • 24
  • 25
  • 26
  • 27

W
HITE
P
APER
(cont.)
17
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
278A/0497
not a big increase in performance since FTP transfers are still doing DNS Lookups. On HTTP
only transfers, the TPM and the percent TPM difference increase from the base system is higher
because the HTTP daemon supports the switch for no DNS Lookups and FTP currently does not.
Please refer to section Test Results with HTTP Only for HTTP only test results.
Turning off HTTP caching of the rules, run #6, degrades performance overall by 1% from the
base system with a peak low of 4%. The overall average failure rate for both tests is less than 1%
from the base system. The performance degradation is expected since caching allows the rule
base only to be checked once a minute as opposed to every connection. Allowing HTTP caching
of the rules especially improves performance if the rule base is large.
MaxRecieve Buffers for NetFlx-3 Cards
Setting the MaxRecieve buffers for the NetFlx-3 NIC cards to 500 represented by Run #7 and
Run #8. Run #7includes the MaxRecieve Buffers change and 2 PCI Netflx-3 cards. This results in
an overall 16% increase from the base system. Run #8 displayed negative differences of 5% from
the base system for 1 to 24 virtual clients and increased as the load increased to an average of
5.6% above the base for 32 to 72 virtual clients. The overall average for all percent differences
for 1 to 72 virtual clients was 1.6%.
Failures decreased from the base system to overall 0.02% for
all virtual client runs.
Processor
In Run #9, a Pentium Pro 200 MHZ-512 cache processor was added to the base system.
Review
Graph 4 below for results.
Graph 4:
Base Run with 2 Processors
Adding the second processor clearly shows performance improvements to the TPM from the base
system.
The overall average percent difference increase from the base system for all virtual
clients is 15%.
As the load increased, the TPM is increased bringing the positive percent
difference to over 36% from the base system with 72 virtual clients.
Failures increased slightly to
2% overall for all virtual clients. The increase in TPM is due to the use of IO Completion ports
and threads in the HTTPD of Raptors Eagle NT 3.06 firewall software.
0
100
200
300
400
500
600
700
800
900
1000
1100
Base Run
2- PP200, 512c
1
12
24
32
36
48
56
72
Virtual Clients
Run #9 in TPM