The OCEANS group at Boston University has done a variety of Web characterization work over the past four years. Our interest in attending the W3C's Workshop on Web Characterization is to interact with other people in the Web characterization community and to present our recent work on HTTP characterization which is titled: A Performance Characterization of Hyper Text Transfer Protocols (abstract below). We have contributed regularly to the HTTP-NG's Workload Characterization Group over the past year and hope to contribute in the future in the areas of workload generators, log file contributions, analysis of specific logs and analysis methods.
Abstract:
Version 1.1 of the Hyper Text Transfer Protocol (HTTP) was developed as a means for reducing the document transfer latency as well as reducing network packet traffic. The mechanisms proposed in HTTP/1.1 assume the network to be the bottleneck in Web transactions, however under heavy load, a Web server can be the primary source of document latency. In this paper, we characterize and evaluate the performance of HTTP/1.0 and HTTP/1.1 in terms of throughput at the server and transfer latency at the client. We use two popular Web servers in our study: Apache v1.3.0 and Microsoft's Internet Information Server (IIS) v4.0. We use the SURGE workload generator on client systems to generate a range of loads on the servers. We present detailed measurements of the server's major subsystems which show that when the CPU is the bottleneck resource then there is relatively little difference in performance between HTTP/1.0 and HTTP/1.1. We show that when memory on the server is constrained, performance using HTTP/1.1 can be much worse than HTTP/1.0. We also show that when the network is the bottleneck resource that performance under HTTP/1.1 is better than HTTP/1.0. Finally, we demonstrate that a simple modification to HTTP/1.1 can greatly improve throughput and latency when memory is constrained.