Randomness Betters Nearest-Rarest in the P2P Clustering Networks

It is crucial for P2P applications to control their traffics for saving the Internet bandwidth. During the P2P communication, if peers tend to share the data with near peers, the Internet traffic can be reduced. Thus the topology of the P2P networks will be clustering. In this article, we study downloading strategies for the living P2P applications with clustering topology. First, we combine two popular downloading strategies: rarest-first and nearest-first. But, there are too many peers always share very few rarest-pieces owned by few peers, so the conflicts happen which will arouse the Internet traffics. We find the randomness policy can better the method by the follows: any peer will request few random pieces from the far neighbors. The goodness is that the near peers will have more rarest-pieces to download. Though such a manipulation produces little additional Internet traffic for a short period, it will decrease the traffic in the long run. The results of our experiments manifest the result and also show that the improved method doesn´t deteriorate the downloading speed.