The disparity between P2P overlays and ISP underlays: issues, existing solutions, and challenges

Jie Dai, Fangming Liu, Bo Li
2010 IEEE Network  
owadays, the Internet has witnessed a remarkable increase in the popularity of peer-to-peer (P2P) applications over ever increasing user bases, which has placed a tremendous traffic burden on the Internet backbone and unprecedented pressure on Internet service providers (ISPs). Reportedly, P2P applications contribute as much as 70 percent of Internet traffic worldwide, and a prediction of 2.8 exabytes of traffic per month will be consumed by those P2P users in 2011, which is enough to make any
more » ... SP cringe. To deal with the considerable traffic costs, many network service providers have started to throttle P2P traffic or limit their bandwidth consumption. This raises the issues and challenges of an ISP-friendly design, which relies on both improvement in P2P applications and assistance from network providers. In P2P applications, individual computers that communicate with each other in a distributed manner to share information and computation, storage, and bandwidth resources can dynamically form self-organized and scalable network topologies, with respect to both virtual overlay network topology at the application level and underlying physical network topology. The mapping between the overlay network topology and underlying physical network topology has a critical impact on not only P2P application performance and service quality, but also ISP traffic management and costs. In particular, the key challenge that has recently received significant attention in both research and practice is that on one hand, many networkoblivious P2P applications without awareness of ISP boundaries can lead to inefficient utilization of Internet resources and poor system performance, as well as resulting in immense costs for ISPs due to the costly inter-ISP traffic; on the other hand, current existing ISP-friendly P2P designs that aim to mitigate inter-ISP traffic by optimizing peering locality could still incur potential defects in violating ISP policies and degrading the robustness and performance of P2P applications. It turns out that the issues associated with the cost, performance, and robustness of P2P applications are highly correlated with the graph-theoretical properties of P2P network topologies, including clustering coefficient and network diameters [1], node degree distribution, and edge reciprocity [2] . Based on large-scale measurement studies involving millions of users, it has been shown that both P2P file sharing networks [3] and P2P video streaming topologies [4] exhibit the small-world effect. That is, peers tend to be highly clustered with small pair-wise shortest path lengths, as compared to a random graph of similar peer population and link densities. This, together with the high-level reciprocity among peers in exchanging data, can facilitate quick and stable data distribution throughout the entire topology [4] . However, to preserve such advantages while further mitigating ISP costs from P2P applications is nontrivial. A recent study showed that peering locality optimization alone could hamper the structural robustness of P2P overlay topologies, due to the potential disruption on crucial connections with relatively low redundancy [5] . In conjunction with assistance from ISPs and collaborative cache infrastructures, how to apply the understanding on the graph properties of P2P networks and theoretic optimization techniques [6] to design and optimize ISP-friendly P2P navigating algorithms is a promising research direction. Challenges The tussle between P2P applications and ISPs over the explosive growth of P2P traffic in the Internet originates from overlay-underlay network oblivion, in which most P2P systems either explicitly or implicitly form an application-layer overlay Abstract The proliferation of peer-to-peer applications has generated tremendous traffic in the Internet backbone and has also posed unprecedented pressure on Internet Service Providers. Many P2P applications, oblivious of underlying ISP networks, can lead to inefficient utilization of Internet resources and a significant amount of costly inter-ISP traffic. Recently, increasing efforts have been directed toward mitigating ISP costs from P2P applications while preserving the user perceived service quality. In this article we review state-of-the-art research on P2P applications with particular focus on their topological properties and navigating algorithms with awareness of ISP costs and performances. We discuss ISP-friendly designs in P2P applications by identifying their respective benefits and deficiencies. Based on the lessons we have learned, we further highlight future research challenges and issues regarding the joint design and optimization of P2P system performance and ISP traffic and costs.
doi:10.1109/mnet.2010.5634441 fatcat:gbxs7h2yfbgf5hhlyiohgpq7vq