Effective Techniques and Tools for Resource Management in Clustered Web Servers

College Of William And Mary, Williamsburg VA

Investigators

Abstract

The explosive growth of the World Wide Web, the variety of information hosted by a given site, the rapid fluctuations in user demand, and the complexity of the preferred architectural trend are among the reasons that make increasingly challenging for system administrators to effectively mange today's complex web server systems. This proposal addresses this problem by seeking to develop smart algorithms that help performing difficult tasks such as web server cluster configuration, capacity planning, fault management and load balancing, by continuously monitoring and adapting to changes in the workload and system. Workload mointoring will provide input to analytic models whose solution will in turn be used by both off-line and on-line algorithms that can propose improved system configurations. To this end, the proposed research objectives will provide: (a) methods to derive detailed statistical workload characterizations of web servers, (b) new and efficient solutions for analytic models of systems that serve tasks drawn from heavy tail probability distributions, (c) a software tool particularly targeted to load balancing in clustered web servers, to be used by non expert modelers, including system administrators, and (d) workload-aware and system-aware algorithms that significantly improve performance and ease of web system management.