OPTIMAL TASK OFFLOADING IN FOG-ENABLED NETWORKS VIA INDEX POLICIES

Abstract

Fog computing has been considered to be a potential solution to enable computation-intensive and latency-critical application at the battery-empowered mobile devices. Task offloading could take advantages of available neighboring computational resources to achieve low latency. By exploiting the temporal correlation of the states at fog nodes, a sequential decision-making problem that aims to optimize the long-term task offloading performance is considered in this paper. Such a problem is a partially observed Markov decision process with an exploitation-exploration tradeoff that is difficult to analyze. We address this tradeoff in task offloading under the framework of restless multi-armed bandits (RMAB). The indexability analysis of this task offloading problem is then provided. Meanwhile, an index policy, which is asymptotically optimal and has remarkably low computation complexity, is established based on the Whittle's index to solve the task offloading problem. Numerical results show the superiority of the proposed task offloading method.