SEECS Colloquium

A common solution in order to meet the need for high availability and fast service on an ever increasing Web site customer population, is clustering with a single system image, i.e. having a set of multiple hosts behave as a single one. This approach has presented new challenges as the performance perceived by each individual user highly depens on both system design and allocation of its resources. Additionally the performance of such systems and the complexity of the means to analyze them is affected considerably by the high variability and rapid fluctuations of their workloads. We propose a new strategy for the allocation of requests in clustered Web servers, based on the size distribution of the requested documents. This strategy manages to achieve a balanced load to each of the back-end servers and its parameters are obtained from analysis of past system data. To study its performance, we use phase-type distribution fittings and solve the resulting models using a new efficient solution method, ETAQA. Performance analysis results show that our size-based policy greatly outperforms random allocation, performs comparably or better than the Join Shortest Weighted Queue and Join Shortest Queue policies, and maximizes cache hits at the back-end servers, therefore behaving similarly to a "locality-aware" allocation policy, but at a very low implementation cost.

Alma Riska is currently completing her PhD in Computer Science under the direction of Prof. Evgeni Smirni at the College of William and Mary. Her interests include computer performance evaluation and its applications to tools for software and system design. She received her undergraduate degree at the University of Tirana (Albania). She has published several papers in journals and conferences such as Performance Evaluation, ACM SIGMETRICS, the TOOLS conference and at other events.