The recent outages and brown-outs at several large Internet portal sites (trading, auction, etc.) and the associated loss in revenues highlight the central role that web-based services play in the E-commerce era. Unpredictable demands posed by sudden surges of requests from users on the open Internet have made it extremely difficult for web site operators to offer predictable quality of service (QoS). In order to provide stable service to users considered important by the site operator under these conditions, it is imperative that systems handle overloads gracefully. In this paper, we describe the design and implementation of a QoS-enabled web server called the Web2K server. The Web2K server enables web site operators to optimize the usage of their web sites by prioritizing among requests from different classes of users. Under overload conditions, predictive queue controllers incorporated in the Web2K server enable it to selectively deny service to some percentage of requests based on the associated user classification. Unlike current web servers, the Web2K server also maintains stable response times under overload using a novel, traffic-aware request processing model. A primary concern of operators while deciding to deploy QoS solutions relates to the overall impact of QoS provisioning on the capacity of their servers. To address this issue we present a comprehensive performance study that quantifies the overheads of a Web2K server and demonstrates its significant benefits. 20 Pages
PDF
download
878987797
028-85220240
