CAPU: Enhancing P2P File Sharing System with Capacity Aware Topology

Abstract

Measurement works show that the unstructured P2P file sharing systems such as Gnutella face the problem of poor scalability and inefficiency search for unpopular items. In this paper, we propose new mechanisms that greatly enhance the performance of file sharing system. Our work exploits the prevalent heterogeneity of the nodes in existing unstructured networks in terms of capacity to construct a quasi-hierarchical topology-aware topology which achieves approximately optimal system throughput. Based on this overlay topology, we propose proactive file index propagation scheme to facilitate search. We also introduce a two-stage search algorithm integrate probabilistic biased random walk that search for popular items and low-redundant multicast (MPR) searching for rare items, achieving approximately O(1) search efficiency for popular items and receivable search latency for rare items respectively. We evaluate our design through simulations and the results show 3 to 5 orders of magnitude improvement in total system capacity compared to other Gnutella-like system.

This work was supported by the National Natural Science Foundation of China under Grant No. 60433040.