HP ProLiant 4500 Communique - Novell BorderManager Performance on Compaq Serve - Page 4

ECG028.0897 Communiqúe (cont.) ... PERFORMANCE RESULTS Proxy Cache Proxy cache tests were conducted using Ziff Davis WebBench™ 1.1 (WebBench™ simulates Web server access). When proxy caching was enabled, the utilization on the Web Server was very low. Web Server access only occurs when requested pages are not cached on the proxy server. As such, a dramatic performance improvement is observed. • Requests per second (RPS) increased by an average of 54.3% with a maximum of 118.6% • Throughput increased by an average of 52.8% with a maximum of 116.8% Note: Throughout the proxy cache tests, the network utilization on one server NIC averaged 40%50% with a peak utilization of 62%. 1. For basic router functionality, a ProLiant 800 with 64MB of RAM is the best choice. Processor utilization averages 5% with no noticeable decrease in performance. A more powerful server is unnecessary. 2. The ProLiant 5000 1P 128MB RAM performed slightly better than the ProLiant 800 1P 128MB RAM. 3. Increased client loads demand additional processing power, therefore a 2-processor ProLiant 5000 is recommended. Utilization rises due to an increase in processes required to service network interface cards. However, please bear in mind that these are preliminary test results, a more detailed analysis with benchmarking will follow later. Note: In a real-world environment, the performance gain will be even higher considering most of the outgoing links to the Internet / Extranet are based on T1 links while these tests were conducted with a 100Mbps link from the Proxy Server to the Web Server. HTTP Accelerator (Reverse Proxy) Unlike a normal proxy configuration, with HTTP acceleration, there is a performance improvement compared to the baseline of 48 clients and above. • Requests per second (RPS) increased by an average of 325% with a maximum of 575% • Throughput increased by an average of 352% with a maximum of 591% • Memory does not affect performance as much as it did under normal proxy configuration for the ProLiant 5000. • Increasing memory has no effect on performance with low client loads but provide minimal increase of performance at higher client loads. • Adding processors has minimal effect on performance with low client loads but provide modest increase of performance at higher client loads. • Performance of a 2-processor configuration increased by an average of 11.1% with a maximum of 38.9% over a 1-processor configuration. However, it should be noted that the processor utilization is 100% on the 1st processor and around 5% on the 2nd processor. The gain comes from servicing the interrupts generated by the network cards in the server. With higher client loads this can become critical. 4