RepliCache : enhancing web caching architecture with the replication of large objects

Abstract

This thesis proposes a RepliCache that scaled well to a large size of objects. Combining caching and replication, RepliCache can amplify advantages of both approaches. Web caching system avoids sending same documents repeatedly over Wide Area Network(WAN) by copying them at a closer location to users. Replication is a popular method for distributing server and network load in the Internet. An analysis of Web workload reveals that increasing fraction of the traffic volumes is associated with requests for large files. Large files such as audio and video data cause heavy traffic both in network bandwidth and data storage due to their large size. The existing Web caching servers, however generally store files no larger than tens of Mbytes to keep small and popular objects. Thus useful network bandwidth is mostly consumed by their unnecessary retransmission and their access latency is large. By storing large objects in a dedicated server and relaying through a proxy cache, large objects can be accessed without a priori knowledge of server location. Moreover, storage can be fully utilized by managing cached objects according to each characteristics. It is also designed ensuring compatibility with existing networking standards, applications and system software.