A Study On Optimization Methods Of Application Deployment In Mobile Edge Computing

With the rapid development and wide applications of mobile and Internet-of-Things(Io T)devices,various high-demand services and applications have continuously emerged.Such type of applications has the higher requirements on network latency and stability than the traditional software.In this case,mobile edge computing(MEC)has attracted the attention from the researchers over the world.In MEC,in order to reduce network latency and improve the overall quality of service,application providers trend to deploy applications on the edge servers nearby users.From the perspective of application deployment,the limited resources of edge servers,user’s requirements for service latency,and the deployment cost of applications are all constraints that need to be considered and satisfied.Therefore,finding an optimal deployment solution that meets the requirements of both the customer and the application provider is a challenging problem.In the existing studies,the application deployment problem is usually divided into two categories of optimization: single-objective optimization and multi-objective optimization.The single-objective optimization model only considers a single objective in the set of constraints,but the real problems often needs to consider multiple objectives.As a result,multi-objective optimization model is more suitable for the practical application scenarios.In this thesis,a Multi-objective Edge Application Deployment(MEAD)problem model is constructed to address the multi-objective requirements of the application deployment problem.The problem model is described as a resource-constrained integer linear programming problem that integrates application provider cost and user access latency constraint,and also considers the sharing ability of applications and connectivity among edge servers.The edge server network is modeled as a graph,where the edge servers are represented by nodes,and the links between edge servers are represented by edges.For the small-scale MEAD problem,this thesis proposes an integer linear programming-based optimization method named MEAD-opt,and uses the linear programming solver CPLEX to solve the problem,and then obtains the optimal solution of the MEAD problem.For the MEAD problem in large-scale scenarios,this thesis proposes a multi-objective application deployment algorithm MO-MEAD based on multi-objective genetic algorithm.MO-MEAD is essentially a meta-heuristic algorithm with the ability to handle large-scale optimization problems,so the algorithm can effectively find the approximate optimal solution.To verify the effectiveness of the proposed algorithms in practical applications,this thesis conducts experimental evaluations with the widely-used public data set and performs the comparison analysis with some other representative algorithms.The experimental results show that the MEAD-opt algorithm has higher efficiency and better solution results compared with the existing baseline algorithms.Meanwhile,the MO-MEAD algorithm can obtain the higher quality solution with lower time consumption than the existing algorithms.In other words,it can provide users with low-latency application services while reducing deployment cost.It can be seen that the MEAD optimization model and the corresponding solving methods proposed in this thesis can provide some guidance and reference to the application deployment problem in MEC.