Asynchronous best-response dynamics for resource allocation games in cloud computing

Abstract

Cloud computing refers to a computing paradigm in which a service provider (i.e. server) provides computing resources to service receivers (i.e. clients) who have heterogeneous requirements for completing their computational jobs. In this paper we consider a decentralized auction-based mechanism for studying this resource allocation problem. This model allows for a general class of queueing processes (representing the computational job arrivals) along with a general class of incentive-compatible bidding mechanisms. This gives us insights into the interplay between the economic and queueing considerations of clients in real-world cloud computing systems. Specifically, we show existence and uniqueness of a Nash equilibrium for the induced game and also show that asynchronous best-response dynamics are sufficient for achieving this equilibrium. The distributed and decentralized dynamics require little communication, thus providing a scheme that can be used to guide implementations in practice.