On P 2 P systems for enterprise content delivery ∗

In this paper we address the problem of optimizing p2p systems for enterprise bandwidth cost savings. Consider a scenario where an enterprise uses p2p swarming to deliver content to its clients. A content may be a file or a package (bundle) of files. For popular and large content, our work finds that enterprise servers can exclusively rely on the cooperation among the peers in the form of a swarm to disseminate the files. We introduce the idea of self sustaining swarms, i.e., swarms where the presence of the server is not necessary for the content to be efficiently transmitted. We characterize, under different scenarios, the concept of self sustainability. We then focus on swarms that are not always self sustaining, and propose a scheme for the server to decide how much bandwidth to allocate to a swarm as a function of the arrival and departure rate of the peers. To show the applicability of our controller, we deploy BitTorrent clients in PlanetLab and in a local cluster. Our emulations confirm that the controller leads to significant improvements in terms of cost savings without degrading the download time for the clients. Finally, we analyze the case of an enterprise providing content through multiple swarms, and present an optimization framework that can be used to determine how much bandwidth should be allocated for each of the swarms.