A Research On Latency Minimization With Shared Data For Multi-user In Mobile Edge Computing

In recent years,a large quantity of intelligent mobile devices have emerged and brought great convenience to our lives.Meanwhile,those devices generate huge volumes of data and have significantly increased the traffic load of the communication networks.Furthermore,the latency requirements in certain latency-sensitive communication scenarios are extremely demanding,which makes it necessary to further reduce the routing delay.As a new communication and computation paradigm,mobile edge computing(MEC)has been proposed to improve the task processing efficiency and relieve the burden of the network traffic by offloading the computing tasks to adjacent edge servers for processing.In this paper,we investigate a multi-user latency minimization with shared data in mobile edge computing.We focus on the task model in some specific scenarios,such as augmented reality(AR).In those scenarios,a single mobile device not only needs to deal with its own individual tasks,but also needs to deal with shared tasks that all mobile devices in the region have to deal with.If shared tasks processing is completed,its computation results can be shared by all the mobile devices in the region.We jointly consider computation offloading and communication to achieve the task offloading strategy.Firstly,we establish the system model,including the task model,the channel model,the task offloading model and the latency calculation model.Then,we formulate the problem as a multi-user multi-task latency minimization problem to improve the quality of experience of mobile devices.The problem is formulated as a mixed integer nonlinear programming(MINLP)problem,which is NP-hard.We adopt two method to solve this problem: convex optimization method and deep reinforcement learning method.For convex optimization method,we first relax the considered problem into a convex optimization problem in our proposed algorithm for ease of computation.We innovatively adopt the successive convex approximation(SCA)techniques to approximate the solution of this relaxed optimization problem in this scenario.Then we project the acquired result to the integer field to acquire sub-optimal solution of original problem.For deep reinforcement learning(DRL)method,we innovatively propose an offloading strategy in MEC based on DRL,which combines the markov decision process(MDP)with the system model in this paper.We adopt deep Q network(DQN)method to solve the problem.And we compare the results with the results solved by convex optimization.Finally,through the experimental simulation,we can acquire that the algorithms proposed in this paper achieve good performance in latency optimization.Moreover,the convex optimization method and DRL method achieve similar results.