Local Area Redirection Policies for a Lightweight Distributed Desktop Webgrid

With the latest innovations and trend towards personalizing users’ web browsing experience, the web has been increasingly dominated by dynamic contents. Solutions abound for improving the performance of static contents today. However, delivering dynamic content remains a challenge due to the many dependencies involved in compiling the content, specifically personalized ones. This paper presents two local area redirection (LARD) policies, the round-robin (RR) and the least busy server first (LBSF) for a cheap, off-the-shelf, local lightweight distributed grid composed of desktop PCs (webgrid). Each LARD algorithm redirects web requests away from a busy server in the webgrid, developed as an alternative architecture for both static and dynamic-content Web sites. The architecture was designed such that performance as characterized by client access time and resource utilizations is optimized during overload conditions, such as flash crowd and time-of-day effects. Through an online simulation technique, we explored the performance savings achieved by the LARD algorithms. Results showed that the end-to-end client delays were improved for both simple and complex jobs. For simple jobs, fewer computers are enough to service the requests at the normal response time. Moreover, the performance continuously improves as the number of computers increases for complex jobs. Results also showed that RR is a better LARD policy than LBSF. LBSF has an added overhead required for the online inventory of the least busy servers in the webgrid.

Keywords –Web servers, request redirection, local area networks, webgrid, desktop PCs, performance evaluation