A Dependable Cluster Based Topology in P2P Networks

Abstract

Unstructured peer-to-peer network is a prevalent model in current P2P networks. In general, unstructured P2P model divides the sharing file into many chunks. A peer must search the positions before downloads a chunk. Since a peer does not know the global topology of the overlay network and the distributions of the wanted chunks, what he can do is flooding chunk queries without a proper order when search and download chunks. Additionally, some peers may behavior selfish: leaves network after completed downloading, downloads but not (or limit) its upload. So, the heavy-tail (or long-tail) phenomenon always exists in the unstructured peer-to-peer networks. Some peers cannot complete their download tasks for the scarcity of certain chunks. By combining topology-control and priority-order, this paper proposes a cluster-based model solve the heavytail problem. It analyzes the distribution of file chunks and increases the replication speed of the rarest chunks transmitted among the group. Our simulation experiments show the proposed model provide a dependent network topology for a P2P system, and it helps to transmit the rarest chunk efficiently and improve the overall download completion rate.