Research On Edge Server Placement Method For System Performance Optimization

In recent years,the rapid development of IoT technology,the increase in the number as well as types of terminal devices,and the development and application of new applications have made mobile devices require a large amount of resources and computing tasks when executing computation-intensive and latency-sensitive applications,and the traditional cloud computing model can no longer meet users’ computing needs.Edge computing can reduce the pressure on cloud computing centers and provide users with low-latency and high-reliability services by introducing edge layer storage and processing computing tasks between users and cloud centers.The issue of edge server placement,as the basis for edge computing environment construction,has received great attention because the location and number of edge servers placed will affect the performance of edge computing systems.On the one hand,efficient server placement can achieve real-time response to users at a lower cost and provide highly reliable and high-quality services.On the other hand,edge server placement plays a decisive role in solving problems such as offloading and resource scheduling in edge computing environments.The existing edge server placement method research has achieved some results,but there are still some problems that need to be further solved,such as not being able to respond to users in time due to the high time complexity of the algorithm and not considering user mobility,etc.This paper addresses these problems and carries out research on edge server placement methods,which mainly includes:(1)For the problem of high time complexity of edge server placement algorithm,a two-stage edge server placement TSESP strategy is proposed.In the first stage,the optimization objective is established to minimize user access delay,balance the load and minimize the cost to build the optimization model,then the exact solution set of a subset of the base station data set is used to build the training set,and finally the classifier is trained with the training set.In the second stage,the edge server placement model ESPML is proposed,and the optimal classifier is selected as the ESPML model by testing.The experimental results show that ESPML achieves near-optimal placement in a short running time and can be used to solve the edge server placement problem in large-scale scenarios.The experimental results on Shanghai Telecom base station dataset and Taiyuan Unicom base station dataset show that the method can achieve better results in terms of access delay and load balancing.(2)For the user mobility problem in the real environment of edge computing,the edge server placement model ESPDQN is proposed.The optimization objectives of this approach are to minimize access delay and balance server load.The edge computing system is defined as a deep reinforcement learning environment from which information about the current server placement state in the system can be observed.ESPDQN continuously optimizes the placement decision and maximizes the reward by continuously interacting with the environment to learn,i.e.,continuously optimizing the performance of the edge computing system and finally determining the server placement decision.Experimental results on two real datasets show that the approach can achieve better results in terms of access delay and load balancing.In this paper,two feasible models TSESP and ESPDQN are designed for the server placement problem in edge computing environments,which optimize the performance of edge computing systems to a certain extent,and the research results further enrich the study of the edge server placement problem.