Dynamic organization schemes for cooperative proxy caching

Abstract

In a generic cooperative caching architecture, web proxies form a mesh network. When a proxy cannot satisfy a request, it forwards the request to the other nodes of the mesh. Since a local cache cannot fulfill the majority of the arriving requests (typical values of the local hit ratio are about 30-50%), the volume of queries diverted to neighboring nodes can substantially grow and may consume considerable amount of system resources. A proxy does not need to cooperate with every node of the mesh due to the following reasons: (i) the traffic characteristics may be highly diverse; (ii) the contents of some nodes may extensively overlap; (iii) the inter-node distance might be too large. Furthermore, organizing N proxies in a mesh topology introduces scalability problems, since the number of queries is of the order of N/sup 2/. Therefore, restricting the number of neighbors for each proxy to k < N - 1 will likely lead to a balanced trade-off between query overhead and hit ratio, provided cooperation is done among useful neighbors. For a number of reasons the selection of useful neighbors is not efficient. An obvious reason is that web access patterns change dynamically. Furthermore, availability of proxies is not always globally known. This paper proposes a set of algorithms that enable proxies to independently explore the network and choose the k most beneficial (according to local criteria) neighbors in a dynamic fashion. The simulation experiments illustrate that the proposed dynamic neighbor reconfiguration schemes significantly reduce the overhead incurred by the mesh topology while yielding higher hit ratios compared to the static approach.