On the Design of Scalable Peer-to-Peer Video Caching

Abstract

Peer-to-Peer(P2P)video caching is a promising approach to accommodate asynchronous requests from cached content at individual peers. However, coherently managing a distributed, heterogeneous, dynamic and potentially large scale cache space is a challenging task. In particular, a key challenge is to effectively control the number of cached copies for popular streams in order to accommodate their concurrent requests with minimum thrashing in the cached content. A few prior studies on P2P video caching rely on the global cache state in order to achieve this goal and, therefore, exhibit limited scalability.

This paper examines key issues in the design of a scalable P2P video caching mechanism that can effectively control the number of cached copies for popular streams by only leveraging the local information at each peer.We argue that the local notion of popularity can serve a sane effective measure to perform cache replacement at individual peers. We sketch two straw-man P2P video caching techniques that rely on the trends in the popularity of individual streams to control the required number of copies in a reactive or proactive fashion. Using simulation, we examine the performance of the proposed mechanisms along with the distributed(and uncoordinated) version of LRU and LFU mechanisms that only use local work load at each peer.Our results show that distributed and uncoordinated P2P video caching generally exhibit good performance across a wide range of scenarios.