Eliminating the Free Rider in Peer to Peer Networks

Eric Chi
(Professor Jean Walrand)
(DARPA) IPTO N66001-00-C-8062

Recent measurements have shown that peer-to-peer file transfer volumes comprise almost half of the traffic carried across metropolitan area networks. What is alarming is that peer-to-peer applications have existed for less than five years. What is even more surprising is that these peer-to-peer communities continue to thrive even though measurements show that only a small fraction of the users are willing to share their content with other users.

In this research, we develop models that predict this unexpected behavior. Users clearly behave in self interest, often at the expense of other users. Cooperation, however, is highly desired since all users will benefit from a more diverse selection of media. Furthermore, the network performance, e.g., lower link loads and shorter download times, should only improve as sharing increases. Moreover, better network performance should only improve a user's utility.

The problem of predicting and driving the behavior of self-interested independent decision makers has been well studied in the context of game theory. We have developed simple repeated game models and found sub game perfect equilibria of cooperation among all users. We allow users to punish each other by either allowing the users to block another user or even anouncing to other users to stay away from a free-riding user. In this way users have an incentive to cooperate.

Questions still remain, however, in making models that will make the desirable equilibrium unique so that under any initial conditions all the users will eventually choose actions leading to an equilibrium of total cooperation.

Send mail to the author : (echi@eecs.berkeley.edu)

Edit this abstract