Popularity based network statistical analysis in peer-to-peer application

Abstract

There are rich statistical characteristics in a peer-to-peer (p2p) network. The more refined statistical characteristics still need further understanding. In this paper we define the popularity threshold of the resource, and abstract the user network based on the popularity threshold to reflect the refined structure characteristics. Through the emprical study of a workload from a dominant peer-to-peer file sharing system, we confirm that the user network based on the popularity threshold has more clear cluster features than the original network. With the popularity threshold of resource increasing, the clustering is more evident. The homoplasy of users within the same cluster is enhanced. The clustering accuracy is inproved. Furthermore, in this paper we extract the cluster fingerprints which can provide a high representation accuracy in low dimensions.