Computation Offloading and Resource Management for Energy and Cost Trade-Offs with Deep Reinforcement Learning in Mobile Edge Computing

Abstract

Mobile edge computing, as a formidable paradigm, sinks the computing and communication resources from the centralized cloud to the edge of networks near to users, which meets the growing demands of mobile applications. However, owing to the mobility of mobile devices and the stochastic edge environment, it is difficult to obtain the prior knowledge of the remaining computing, communication resources and the queuing status of mobile devices and edge servers, how to achieve effective computation offloading and resource management to achieve the trade-offs between completion delay and energy consumption without knowing the prior knowledge of the dynamic edge environment is still facing a great challenge. For addressing this challenge, a computation offloading and resource management method, named PCORA, is proposed for the dynamic mobile edge computing environment. Technically, considering the time-varying characteristics of the network and the dynamic resource requests, the dynamic computation offloading and resource management problem is constructed as a Markov decision process (MDP). Then, the proximal policy optimization (PPO) is employed to obtain the optimal computation offloading and resource management strategy to optimize the completion delay and energy consumption. Finally, comprehensive experimental results are analyzed to illustrate the effectiveness of PCORA.