Strategic Properties of Peer-to-Peer Data Structures and Behaviour of Human Peers in Economic Experiments

Abstract

Structured peer-to-peer networks manage large data sets. Each peer administers a part of the data. To answer a query, several peers must collaborate. However, experience with peer-to-peer file sharing systems shows that peers tend to abandon collaboration. Such behaviour is also expected in structured peer-to-peer systems. In the recent past, protocols for such systems to counter this kind of behaviour have been proposed. Our objective is to investigate how the behaviour of peers reacts to such protocols. We use idealized game theoretic models to analyze the strategic properties of structured peer-to-peer networks. The models predict cutoff strategies. I.e., if more than a certain percentage of their queries are answered, peers also answer queries in a stationary equilibrium. Another theoretical result is that free riding may occur in equilibrium. To check whether our theoretical predictions are in line with reality, we have implemented our model in an experimental laboratory and have analyzed the behaviour of human peers. These experiments support our theoretical results.