Managing flash crowds on the Internet

Abstract

A flash crowd is a surge in traffic to a particular Web site that causes the site to be virtually unreachable. We present a model of flash crowd events and evaluate the performance of various multilevel caching techniques suitable for managing these events. By using well-dispersed caches and with judicious choice of replacement algorithms we show reductions in client response times by as much as a factor of 25. We also show that these caches eliminate the server and network hot spots by distributing the load over the entire network.