Compaq ProLiant 1000 Performance Analysis and Tuning of Raptor's Eagle NT 3.06 - Page 17
MaxRecieve Buffers for NetFlx-3 Cards, Processor
View all Compaq ProLiant 1000 manuals
Add to My Manuals
Save this manual to your list of manuals |
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