P 2 P Infrastructure for Content Distribution

The explosive growth of the Web requires new solutions for content distribution that meets the requirements of scalability, performance and robustness. At the same time, Web 2.0 has fostered participation and collaboration among users and has shed light on Peer-to-Peer (P2P) systems which involve resource sharing and decentralized collaboration. This thesis aims at building a low-cost infrastructure for content distribution based on P2P systems. However, this is extremely challenging given the dynamic and autonomous behavior of peers as well as the locality-unaware nature of P2P overlay networks. In the first stage, we focus on P2P file sharing as a first effort to build a basic infrastructure with loose requirements. We address the problem of bandwidth consumption from two angles: search inefficiency and long-distance file transfers. Our solution Locaware leverages inherent properties of P2P file sharing; it performs locality-aware index caching and supports keyword queries which are the most common in this context. In the second stage, we elaborate a P2P CDN infrastructure, which enables any popular and under-provisioned website to distribute its content with the help of its community of interested users. To efficiently route queries and serve content, Flower-CDN infrastructure intelligently combines different types of overlays with gossip protocols while exploiting peer interests and localities. PetalUp-CDN brings scalability and adaptability under massive and variable scales while the maintenance protocols provide high robustness under churn. We evaluate our solutions through extensive simulations and the results show acceptable overhead and excellent performance, in terms of hit ratio and response times.