Abstract Managing Flash Crowds On The Internet

A flash crowd is a surge in traffic to a particular Web site that causes the site to be virtually unreachable. We present a theoretical model of flash crowd events and evaluate the performance of various multilevel caching techniques suitable for managing these events. Our results show that orders of magnitude savings can be achieved in client response times and server and network load during flash crowds using basic caching techniques, and even more savings can be realized with a bit more intelligence in replacement algorithms and placement of proxies