Quantitative Measurement of High Performance L5-Dispatcher

This paper is mainly concerned with the performance evaluation of HTTP Splicing. Aiming at the flaws of TCP Splicing, HTTP Splicing, a method adopted widely by L5-dispatcher which is a sort of web content switch, is put forward. The testing methods include overhead testing, performance testing, testing contrasted with L4-dispatcher, testing contrasted with TCP Splicing, and scalability testing. The conclusions are: when accessing a 1KB file under best effort model, performance of HTTP Splicing is better than that of TCP Splicing at least by 3.5; while accessing a 10KB files under best effort model, performance of HTTP Splicing improves that of TCP Splicing by 22%. When accessing a 307 bytes file under best effort model, the best performance of an implementation of HTTP Splicing, WebPatcher, is 5556.8 replies/s, outdoing ArrowPoint Co.’s CS-100（1193 replies/s） and Alteon Co.’s ACEdirector 3（v6.0.25）（1356 replies/s）.