Research On Key Technologies For Microservice Architecture Application Deployment In Edge Computing

Edge computing,as a novel computational paradigm,offers effective solutions to challenges such as high data transfer latency and extensive network bandwidth consumption.Edge computing is particularly adept at supporting Internet of Things(IoT)applications that require stringent latency requirements,leading to its widespread adoption across various industries.With the increasing complexity of applications and the evolution of development paradigms,microservices architecture,characterized by modularity and loose coupling,is well-suited for deployment on resource-constrained edge devices.Deploying microservices in edge computing environments offers benefits such as rapid request-response times,moderate resource demands,independent service deployment and scaling,and flexible migration capabilities,making it an effective technical approach for enhancing deployment service quality while optimizing overheads.Nevertheless,given the distributed,heterogeneous,and resourceconstrained nature of edge devices,efficient and stable deployment in edge environments has become a focal point for both academia and industry.A microservice architecture aimed at edge computing primarily prioritizes optimizing application request response times.With the goal of maintaining the quality of service for end-user requests,minimizing the resource consumption of application deployment is crucial for optimizing resource utilization in edge computing.Collaborating between terminals and edge computational capacities,judicious partitioning of request data,and efficient computational allocation are paramount to further reducing energy consumption in edge computing.Based on these considerations,this dissertation delves deeply into the deployment of microservice architecture applications in edge computing environments,offering systematic investigations in three areas: response time optimization,resource footprint reduction,and energy consumption minimization,striving for more efficient and stable edge computing services.The research contents are summarized as follows:1.Response Time Optimization in Application Deployment: Existing research often overlooks the influence of varying interaction frequencies between microservices and the performance degradation due to increased node loads.To address this,this dissertation first models the relationships between microservices as an undirected weighted interaction graph,representing communication overheads.Subsequently,a multi-objective microservice deployment model for edge computing is proposed.This model aims to minimize communication costs while ensuring load balancing across edge nodes.Using deep reinforcement learning,this dissertation derives optimal deployment strategies without the need for domain-specific expertise.Additionally,a heuristic-based scaling algorithm is introduced for handling dynamic request pressures.Experimental evaluations on Kubernetes demonstrate that our approach offers shorter response times,balanced resource loads,and elastic service scaling in response to varying request loads,outperforming interaction-aware strategies and default Kubernetes policies.2.Resource Consumption Optimization in Application Deployment: Much of the existing research disregards applications with multiple call graphs and varying dependencies,often resulting in the breach of service quality for some call graphs.To tackle this,this dissertation meticulously model request response times for scenarios with multiple service instances,call graphs,and service conflicts.Our objective is to minimize the number of deployed service instances,thus reducing resource consumption,while upholding the service quality constraints of all call graphs.Capitalizing on the non-linear representation capabilities of Graph Neural Networks for multiple call graph structures,this dissertation proposes a graph-reinforcement learning-based deployment framework.The framework utilizes graph convolution networks to extract features of multiple call graphs and feeds these features into a deep reinforcement learning network.Iterative agent-environment interactions yield an optimal deployment sequence.Experiments reveal that our method effectively diminishes the number of deployed service instances,reducing resource consumption,all the while meeting service quality constraints.3.Energy Consumption Optimization in Application Deployment: Given the enhanced computational capabilities of modern end devices,it is feasible to deploy certain microservices locally to optimize service quality.Offloading part of the request data to edge servers creates a synergy between terminal and edge computational capabilities.Determining the optimal offload ratio and the computational resource allocation on edge servers becomes crucial.Current offloading strategies,focused solely on latency minimization,often offload excessive data,leading to significant energy overheads.A Quality of Service(QoS)-aware joint computational offloading strategy is presented,aiming to minimize overall energy overhead while avoiding resource over-allocation and high energy consumption.A Bayesian optimization-based method is introduced to determine the best data offloading and computational resource allocation decisions.Experimental results highlight the method’s ability to ensure service quality while outperforming existing algorithms in energy overhead reduction.Building on these studies,related technologies are applied to scheduling optimization problems in orbit edge computing from the perspectives of on-orbit resource management and task scheduling engineering frameworks,as well as cooperative algorithm optimization between high and low orbit satellites.In summary,this research provides a comprehensive analysis and algorithmic design for deploying microservices architecture applications in edge computing environments,focusing on response time,resource utilization,and energy consumption optimizations.Both theoretical and numerical experiments validate the feasibility and superiority of the proposed methods.