Learning based utility maximization for multi-resource management

Abstract

The future mobile infrastructure is expected to become a multi-resource environment where various re-sources participate in user's services. Moreover, as users' needs become diversied, network managerrequires algorithms to allocate resources eciently and fairly. However managing multi-resource envi-ronment is challenge due to time varying nature and complex correlation among service requirement andexisting algorithm have shown poor performance in metric like queuing delay. In this paper, we aim toovercome these limitations through learning based algorithm. We modeled the multi-resource manage-ment problem in multi-resource environment with computing/networking resources as utility maximiza-tion problem, and reformulate the problem to appropriate form to apply the reinforcement learning. Asa result, the algorithm we proposed can achieve utility optimal without any trade-o parameter and doesnot cause high queuing delay unlike existing algorithms.