Performance study of satellite-linked web caches and filtering policies

The exponential growth of World-Wide Web (WWW) requires new and more efficient content distribution mechanisms. Web caching has shown to be a very efficient way to scale the WWW by reducing traffic in congested network links, decreasing latency to the clients, and reducing load in the origin servers. In this paper we study an emerging caching strategy called cache-satellite distribution. In a cache-satellite distribution caches are inter-connected via a satellite channel, effectively increasing the client population connected to a cache. Using Markov chain analysis, we demonstrate that the higher the number of clients connected to a cache-satellite distribution, the higher the probability that a document request is hit in the cache. Our analytical results strongly advocate a large scale interconnection and cooperation of ISP-caches via a satellite distribution. Due to the large amount of documents distributed through the satellite and the limited disk capacity and processing power of local ISP caches, it is impossible to store all documents coming from the satellite distribution. Therefore, it becomes necessary for an ISP-cache connected to the satellite distribution to perform efficient filtering policies. Based on the stability in the behavior of the clients of an ISP cache, we propose novel filtering policies which rely on the Web servers visited on the previous days. Using trace-driven simulation, we study different filtering policies and show how simple filtering policies can exhibit excellent performance in rejecting non-desired documents while assuring high hit rates.