Research On Microservice Deployment And Composition Optimization In Mobile Edge Computing

Mobile/Multi-Access Edge Computing(MEC)is an emerging distributed computing paradigm that effectively improves user service experience by pushing computing capabilities to the edge nodes.The lightweight communication mechanism of microservices architecture is widely used in application development in the MEC environment to achieve flexibility and scalability of applications.However,in the MEC environment,the deployment and composition of microservices face problems such as edge server’s location dispersion,server’s resource limitation,edge device heterogeneity,and environmental dynamics.Suitable techniques and methods need to be adopted to address these issues and achieve efficient,reliable,and scalable microservices deployment and composition solutions.In the MEC environment,how to optimize the deployment and composition of microservices to maximize system performance and user experience is a problem worthy of in-depth research.Through theoretical analysis and experimental research,this dissertation has achieved the following main research results.(1)In response to the complexity of deploying edge servers in the MEC distributed architecture,a Glowworm Swarm Optimization Edge Server Placement(GSOESP)method is proposed under the goal of minimizing service access delay and load differences among edge servers in the MEC environment.Based on relevant theoretical analysis,edge server deployment in MEC environment is a multi-constraint,multi-objective optimization problem.Considering factors such as base station distribution,communication latency,user request distribution,and edge node computing and storage capabilities,a deployment optimization model for edge servers is established,combined with the practical requirements of microservice deployment and combinatorial optimization.Based on the improved firefly algorithm,the GSOESP method is proposed to find the optimal deployment location of edge servers to minimize user service access latency.At the same time,according to the load priority and gradually approaching the optimal target principle,each base station is assigned to an edge server to solve the problem of uneven server load.To evaluate the effectiveness of the GSOESP method,performance tests were conducted based on a real dataset from Shanghai Telecom.Experimental results show that the GSOESP method achieves a balance between "low latency" and "load balancing" while ensuring user service access quality,and its effect is better than the other three existing comparison methods.(2)To address the challenges posed by limited resources and heterogeneity of edge devices in the MEC environment,which present challenges for microservice deployment,an Adaptive dynamic deployment optimization method(Adapt-SD)is proposed under the goal of minimizing user service access costs through self-adaptive and dynamic deployment of microservices.Based on the state information such as the amount of service requests from mobile users,the capacity and processing ability of edge servers,the microservice deployment problem in the MEC environment is modeled as a Microservice Deployment Optimization Problem(MSDOP).Based on the Adam optimization algorithm and weighted round-robin scheduling algorithm in deep learning,the Adapt-SD method is proposed to solve the MSDOP problem.To evaluate the effectiveness of the Adapt-SD method,performance tests were conducted based on a real dataset from the EUA region of Australia.Experimental results show that the Adapt-SD method achieves the minimum resource cost for user access to services while maximizing the satisfaction of user service access latency constraints,and its effect is better than the other four existing comparison methods.(3)In order to address the challenges posed by the mutual constraints of service evaluation indicators and the dynamism of service requests in the MEC environment,which present certain challenges for microservice composition,a Microservice Composition based Multi-Objective Evolutionary(MSCMOE)optimization method is proposed under the goal of minimizing service access delay and resource consumption through joint perception of delay and cost in the MEC environment.Combining with the relevant characteristics of microservice composition problems in mobile edge environments,the microservice composition process is represented as a directed acyclic graph(DAG)with dependencies between microservices,constructing a multi-objective optimization problem for microservice composition.The MSCMOE method,which is based on an improved non-dominated sorting genetic algorithm,is used to solve the microservice composition problem.While maintaining population diversity,the computational efficiency of seeking elite solutions in the non-dominated layer is improved by using an improved reference point strategy.To evaluate the effectiveness of the MSCMOE method,performance tests are conducted using real datasets from Shanghai Telecom.Experimental results show that the MSCMOE method achieves the goal of minimizing network resource consumption while reducing user service access latency,and its performance is superior to that of four other existing composition methods while also having good convergence.(4)Aiming at the problems of service interruption or delay caused by continuous user mobility in MEC environment and with the goal of minimizing user service access delay,a microservice composition optimization method based on user mobility awareness is proposed,named MS＿DDPG(Microservice composition based Deep Deterministic Policy Gradient).Based on the key factors such as user mobility and edge server hardware resource constraints,the mobility-aware microservice composition optimization problem(MSCOP)is studied.The MSCOP problem is modeled as a Markov decision process,and the microservice selection policy experience pool is introduced to cope with the complexity of mobile edge environment and improve the learning efficiency.It is proven that the optimal or suboptimal microservice composition strategy can be obtained for the MSCOP problem.Based on deep reinforcement learning algorithm,the MS＿DDPG method is proposed to solve the MSCOP problem.To evaluate the effectiveness of the MS＿DDPG method,the real data set of EUA in Australia is used for performance testing,and the convergence speed of the MS＿DDPG method is also evaluated.The experimental results show that MS＿DDPG outperforms other three existing microservice composition methods in terms of user service access delay,and it has good robustness as well.