Fairness, incentives and performance in peer-to-peer networks

In this paper we consider the sensitivity of the ‘service capacity’ of peer to peer (P2P) file sharing applications to demands, user cooperation, and providing incentives for cooperation through fairness guarantees. To do so we propose models for both the steady state performance and the transient characteristics of the system. Both are of interest since, from the perspective of a given file, demands are prone to be initially bursty, e.g., flash crowds, and subsequently be roughly ‘stationary.’ We will show that P2P systems can be exceptionally responsive to bursty demands by increasing their service capacity exponentially, while exhibiting nice performance scaling to the offered loads during stationary regimes. We also study system characteristics associated with various fairness criteria. We provide partial validation of our models and conclusions based on traces obtained from a second generation P2P application.