Popularity-based scalable peer-to-peer topology growth

Abstract

Peer-to-peer (P2P) networks have gained importance and spread significantly during the last decades resulting in raised research interest in this area. The basic premise of P2P design is higher scalability and many existing large-scale applications, such as Twitter and Skype, use a form of P2P design. Although structured P2P designs enable one to guarantee finding of every item (rare or popular), they do not scale beyond a point and support from servers are needed. This breaks the decentralized design of the P2P system and results in a hybrid scheme. Unstructured P2P networks, on the other hand, can scale to much larger nodes but yet cannot give time guarantee for finding a rare item.Topological characteristics of unstructured P2P networks have impact on the efficiency of a search for items. Earlier studies have shown that popularity of the nodes has significant impact in the self-organization of the overlay topology. It is well known that unstructured P2P designs are very good when searching for a popular item. But, when a new trend is emerging, it is a rare item until it becomes highly popular. During that transitional period, unstructured designs suffer from inefficiency in finding to-be-popular (TBP) items. Such TBP items could be stuck at nodes with small degrees and thus become hard to find. We hypothesize that if the overlay topology is established and grown with a more proactive consideration of the items’ popularity, the delay in finding TBP items could be reduced significantly. Further, such topology growth will reduce the search time for popular items. Thus, the overall search performance of the P2P system will significantly improve as well, since most of the searches are for popular items. In this paper, we investigate incorporating item popularity into the overlay topology in a scalable way.