Compaq ProLiant 1000 Performance Analysis and Tuning of Raptor's Eagle NT 3.06 - Page 20

278A/0497 WHITE PAPER (cont.) ... Other Systems and Configurations This section includes test runs with other systems and configurations. It also includes other runs using the base system described above with other configurations. Graph 7 includes runs #16, #17, and #18 for ProSignia 500, ProLiant 800, and the ProLiant 4500 respectively. These runs should not be compared to the base system because of the configuration and hardware differences. Descriptions of these three machines and their configurations are listed in thesection Test Configurations of the Firewall. Runs #16, #17, and #18 in TPM 1100 1000 900 800 700 600 500 400 300 200 100 0 Prosignia 500, P120 Proliant 800, PP200, 256C Proliant 4500, P133 1 12 24 32 36 48 56 72 Virtual Clients Graph 7: ProSignia 500, ProLiant 800, ProLiant 4500 The low-end server, the ProSignia 500, had an overall average of 273 TPM for all virtual clients pushing almost an estimated 1.9Mb per seconds through the system. The ProLiant 800 had an overall average for all virtual clients of 620 TPM with a high of 665 TPM at 48 virtual clients. The ProLiant 4500 had an overall average of 398 TPM for its virtual client runs. The average failure rates for these test runs are less than 1%. Another test run on the base system showed that the firewall could handle more than 10Mb of traffic through the firewall for HTTP and FTP transfers. This test added the number of virtual clients to a total of 144 virtual clients on 8 physical client machines. The test results show that the TPM was 1696 and 10.67Mb throughput was calculated for data passing through the firewall. Tests Results with HTTP Only The same model is used for HTTP only sets of tests as for the tests with HTTP and FTP. The same base system is used with no changes except that 100 percent of the traffic is HTTP traffic. The benchmark is run using the base system with the same test bed setup, 3 physical servers and 8 physical clients, with 1 to 72 virtual clients and 6 virtual servers. The set of tests run in this section were designed to show the performance increases that had been made to the HTTP daemon. These increases in performance from the HTTP daemon are attributed to the threaded enhancement, the use of IO completion ports, and caching of the rules. This section presents the Base Run, DNS Lookups for HTTPD and HTTP Cache, and 100 Rules. 20