Research On Application Scheduling And Auto-scaling Optimization Based On Microservice Architecture

As the core technology provided by IoT platform services,microservices support diversified Io T applications.Because the large number of microservices and complex interactions,the deployment and application of microservices for complex business systems are facing severe challenges.Microservice deployment for virtual machine cluster nodes,workflow scheduling to ensure load balancing,and auto-scaling of service provision affect economic costs and resource utilization.With Microthings applications as the background,aim at reduce economic costs and improving server resource utilization,this article focuses on application scheduling and auto-scaling methods under microservices,and proposes a method TSAS based on the combination of task scheduling and auto-scaling,Realize the multi-objective optimization of s economic costs and server resource utilization rate under the micro-service architecture.Aiming at the problem of inaccurate microservice flow scheduling caused by the dynamics of the application structure and service input or output under the microservice architecture,under the microservice flow deadline,taking the economic costs as the optimization goal,a microservice scheduling and auto-scaling optimization TSAS is proposed.The microservice flow is modeled,and a scheduling algorithm based on the urgency of the microservice flow is proposed.By assigning the sub-deadline to the service in the microservice flow,an optimized service execution mapping relationship and the adaptive configuration resource of the container are given.Experiments on different workflows Montage and Cybershake show that the TSAS saves 15.62% of economic costs compared to similar solutions(ESMA,IC-PCP)on average.Aiming at the optimization problem of CPU and memory utilization of virtual machine cluster in the two-layer mode of container and virtual machine,a FSW scheduling algorithm based on greedy algorithm is proposed.The best fit(BF)strategy is set as the number of images required to start the matching container in the virtual machine,and its value is proportional to the BF rate.The eigenvectors of containers and virtual machines are standardized by CPU and memory.and the profit function of each allocation decision is described as a capacity vector through modeling,which contains the CPU usage rate and memory usage rate of the virtual machine cluster,and the container scheduling problem is summarized for the bounded,multi-dimensional and multi-objective knapsack problem.Through three processes such as filtering,generating Pareto solutions,and choosing,the container is allocated to the selected virtual machine.In order to prevent the solution set from entering the local optimal solution,the NSGA-II is introduced to generate the Pareto solution of the container scheduling.Compared with the common container scheduling algorithms Binpack and Spread,it shows that the FSW can save 9.99% of the virtual machine node usage.At the same time,the CPU utilization of the virtual machine cluster increases by 43.1% on average,and the memory utilization increases by 56.1% on average.And the Pareto solution set is better than SPEA2.Based on the proposed TSAS optimization method and FSW scheduling algorithm,the comparison verification is carried out on the workflow platform.The results show that the method proposed in this paper can save economic costs and improve the utilization of virtual machine clusters under time constraints.This method has been integrated into the built Kubernetes to perform service scheduling and horizontal scaling of the services deployed in the cluster,which verifies the availability and feasibility of the researched technologies and solutions.