Implementing cooperative caching in distributed streaming media server clusters

Abstract

In distributed streaming media server clusters, by adopting cooperative caching (CC) technique, the free memory of all the servers can be combined to form a bigger, logically integral cooperative cache. It will help raise the hit rate of the cache and reduce disk accesses, resulting in the improvement of the overall throughput of server systems. Traditional CC and streaming buffer replacement algorithms are not quite suitable for streaming server clusters. This paper proposes a new cooperative caching strategy for them, called GLNU (Globally Longest-Not-to-be-Used). It is based on server level cooperation and fully takes the streaming media's continuous playback requirement into consideration. Compared with traditional CC algorithms and cache algorithms for continuous media, this algorithm is more pertinent and suitable to the distributed streaming server cluster systems. Simulation results show that GLNU has better performance than several other traditional cache algorithms in various conditions. © Springer-Verlag Berlin Heidelberg 2004.