Cloud networking mean field games
Résumé
In this paper we analyze a distributed resource sharing problem for cloud networking. Each user would like to maximize a given payoff based on its demand and the total demand on the cloud. The problem is formulated as a game where the action of each player is represented by its requested demand. We develop a distributed algorithm for each node which only requires mean demand from the cloud to update its respective demand, thus reducing overhead. We also prove the convergence of our algorithm to Nash equilibrium. For large scale systems, we analyze the performance for "selfish" and "social" user strategies with symmetric price, and present a non feedback based distributed algorithm. We compare the performance of our algorithm with existing algorithms. Finally we present numerical results which compares the convergence of feedback vs non feedback algorithms.