Impact of P2P-Assisted and Interaction-Oriented Content Distribution for Online Social Networks

Abstract

Rapid growth in online social networks (OSNs) like Facebook, Twitter, and Flickr draws great attention due to their unique load characteristic and scalability challenges. Unlike traditional websites, traffic from online social networks tends to be tailored and dynamically generated for each individual user. Employing P2P solution to alleviate this scalability problem requires additional investigation due to special social requirement in user experience and privacy control. This paper proposes a hybrid P2P-assisted and interaction-oriented content distribution called PAIRCD that transparently serves OSNs users primarily through P2P data delivery but with the help of central servers. PAIRCD conducts efficient P2P data delivery using social connections and performs optimized content prefetching based on measured interactions among users. Choosing interaction as the criteria can greatly increase system efficiency because user's interaction graph is measured to be smaller than the social graph. Both our theoretical model and extensive experiments show that PAIRCD ensures satisfactory user experience with acceptable overhead on clients' network. What's more, our experiments confirm that PAIRCD achieves one order of magnitude of load reduction for OSNs central servers.