Distributed Dynamic Resource Management In Mobile Edge Computing

Driven by the ever-increasing demands of mobile applications,Mobile Edge Computing(MEC)is envisioned as a promising paradigm for pushing computing resources to the edge of networks.However,how to achieve distributed resource management of MEC under a dynamic environment is still challenging.Based on Unmanned Aerial Vehicle(UAV)-enabled MEC networks,this article conducts in-depth research on the simultaneous realization of efficient computing offloading,server deployment,and UAV trajectory control.First,for the dynamic computing offloading and server deployment in the UAV-enabled MEC network,we consider the random generation of user computing tasks,time-varying channels,and the offline possibility of UAVs,and propose a distributed resource management framework.By formulating multi-user computing offloading and multi-edge server deployment stochastic games,we decompose the system overhead optimization problem into two sub-problems.For each stochastic game,we propose a learning algorithm based on the probability of strategy selection to achieve a pure Nash equilibrium and then incorporate the two learning algorithms into a chess-like asynchronous updating algorithm to minimize the system-wide computational cost under a dynamic environment.Then,we focus on the differentiated services provided by UAVs from different service providers and the time-varying service preferences of ground users in the UAV-enabled MEC network,and propose a deep reinforcement learning-based UAV trajectory optimization method,which firstly formulates the interactions among UAVs as a Markov game,and then constructs each service provider as an agent.Based on multi-agent deep reinforcement learning and prioritized experience replay,the proposed method realizes the distributed trajectory control of UAVs,where each UAV executes flight actions only based on local observations,thereby minimizing the computational cost of all ground users and the long-term computational cost of UAVs,simultaneously.Finally,simulation results based on real-world data demonstrate the efficiency of our proposed distributed resource management scheme,which can greatly improve the system performance of UAV-enabled MEC networks under a dynamic environment.