A study of the impact of structure on cooperation in networks

The concern of this paper is that nodes in a distributed system cooperate. We demonstrate that adding structure to such systems can help to induce cooperation. ‘Adding structure’ means bounding the number of nodes a node can directly interact with - if two unconnected nodes wish to interact, they have to do so via a chain of intermediaries which are connected. To validate our claim, we analyze one specific system where structure has been added. Different methods exist for such analyses: Economists use behavioral experiments to study human behavior. Computer scientists rely on simulations, to obtain results that are robust and scalable. Both use game theory to arrive at formal results. Each approach in isolation has its disadvantages: Behavioral experiments do not allow analyzing many different combinations of parameter values and do not scale well. Game theory supports analyzing system equilibria, but may result in complex or even unsolvable models, in particular when considering system dynamics. Simulations rely on the intuition of the system designer - this is hazardous in strategic settings. We pursue a combination of these approaches and show that the degree of cooperation in our setting is high. More specifically, after having shown with behavioral experiments that humans tend to use cut-off strategies, we translate the strategies used into a game theoretic model and show that they yield an efficient equilibrium. Simulations show that the system reaches this equilibrium given random start populations if the rate of free riders is low. We conclude by saying how our work affects the design of distributed systems from a computer science perspective.