Understanding the performance of offline download in real p2p networks

P2P file-sharing systems have been suffering from file unavailability or poor download speed due to lack of enough seeds. To download rare files, users are typically forced to keep their desktops online for a long time, which is undesirable for most users. To get over this, several commercial P2P systems launched offline download service, also known as cloud download, which employs stable and high-capacity servers to take over users’ download tasks, and thus users do not need to stay online all the time. This paper focuses on characterizing the performance of offline download, including the overall performance provision and the expected performance of individual download tasks. Through model characterization and experimental verification based on real-world trace, we are the first to confirm the feasibility and effectiveness of offline download for a large spectrum of P2P systems: over 90 % files can be successfully retrieved, and most of the downloads only take a few days. Furthermore, we disclose that the completion time of offline download is extremely volatile, and thus develop a method to accurately predict the expected performance of individual downloads. An instance of applying the prediction results is also presented to save the server-side bandwidth. Finally, we implement a prototype of offline download service, release it in a real P2P network, and present our work experience with it.