Alleviating Hotspots with Replication and Redirection Ryan Ungaretti ECE Dept., Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA 15213 +1 412 862 3727 rju@andrew.cmu.edu ABSTRACT Internet hotspots occur quickly and can often cause service outages for web servers. Current solutions to this problem, such as Akamai[1], are not financially feasible to implement for many companies and especially for individual users. This paper presents a simple web server that can participate in a distributed peer-to-peer network, monitor its own bandwidth usage and automatically replicate content onto peer web servers to reduce server load. The server can then probabilistically redirect client requests, using standard HTTP "302 Found" replies, to the peer web servers to access the replicated content. This allows a server that is handling a large amount of requests to redirect them to other peers and reduce the amount of bandwidth being utilized. Because the server replies to clients with redirect messages instead of content, fewer bytes are transferred across the network path and the average response time between the affected server and the client drops. The client suffers only a minor added delay when retrieving replicated content off of a peer server.