User data mining in a large-scale peer-assisted offline download system

Peer-assisted offline download service provides a new Internet content delivery approach, in which the system coordinate dedicated servers and peers to finish a file download. This service has attracted a large number of subscribers. The related research, however, is still missing. This paper presents our preliminary measurement and analysis of users' behavior in QQXuanfeng, one of the most popular peer-assisted offline downloading applications in China. Based on a large set of real- world user data, we validate the performance of system, and cluster users according to their different usage features. The found patterns can be used to exploit potential system strategies on future design. Our work is not only a measurement result of specified goal, but also a new assessment method which can be widely used.