K. Kim and D. Park (Korea)
peer-to-peer, clustering, web caching, replacement algo rithm
Many web cache systems and policies have been proposed. These studies, however, consider large sized objects less useful than small sized objects for the performance and evict them as soon as possible. Even if this approach in creases the hit rate, the byte hit rate decreases and more bandwidth is used to obtain large sized objects. This paper suggests the web cache system which uses the client-cluster which is composed of the residual re sources of clients as an exclusive storage for large sized objects. To manage the client-cluster, we use DHT based peer-to-peer lookup protocol which makes the system self organizing, fault-tolerant, well-balanced and scalable. We manage the large sized object by the index based allocation method and balance the loads of all clients well. Conse quently, this proposed system achieves not only the high hit rate but also the high byte hit rate, and reduces the out going traffics. We examine the performance of the cache system via a trace driven simulation and demonstrate effective en hancement of the proxy cache performance.
Important Links:
Go Back
