The phenomenal growth of the World Wide Web has sparked a lot of research activity on the performance and scalability of Internet Web servers. Many researchers are studying caching architectures for Web documents, Web proxies, and stream-lined protocols for client-server Web interaction.
Our main contribution to date has been a detailed workload characterization study (conducted by Martin Arlitt and Carey Williamson) of Internet Web servers, reported at the 1996 ACM SIGMETRICS conference. This study analyzed Web server access logs from six different sites on the Internet (Calgary, ClarkNet, NASA, NCSA, Saskatchewan, and Waterloo), and tried to identify workload invariants: workload characteristics that were common to all servers, regardless of the level of activity of the time duration of the trace. A total of ten invariants were identified. These invariants were then used to suggest promising approaches for document caching for Internet Web servers.
Our current efforts focus on studying document caching strategies for Web documents, not only for single-level caching at one Web server, but also for multi-level caching, and for Web proxy servers. We are also investigating the design of a high-performance, scalable, distributed Web server. Most of this work is currently proceeding using trace-driven simulations to explore the Web server design space, with implementation of a prototype Web server planned for the near future.
 Research
Areas |
 DISCUS
|
 Research
Groups |
 Comp
Sci. Dept |
 University |