Research On Optimization Of Resource Scheduling Based On Microservice Architecture Application Platform

Since Martin Fowler formally proposed the concept of microservices in 2014,microservices have attracted widespread attention in the industry.The implementation of service architecture has gained sufficient industrial production experience,and microservice architecture is gradually becoming the mainstream of enterprise application architecture.As a style of distributed system development,mcroservice architecture inevitably treats resource allocation and network service as application goals.Due to the characteristics of the microservice architecture itself,resources are wasted due to the division of too many services,so resource optimization has become the main research direction of the current microservice architecture.At present,in the method of resource allocation based on microservices applications,the mainstream methods are to formulate resource allocation strategies based on predicted workloads,optimize resource usage by monitoring application resource consumption to adjust the number of virtual machines,and set rules.Solve the suboptimal solution to formulate the resource allocation strategy,and set the rule to solve the optimal solution to formulate the resource allocation strategy.These methods do not take into account that the fundamental reason for allocating resources is to meet user access needs and save costs as much as possible,so there is still much room for improvement in resource allocation methods.In microservice architecture-based application platforms,different users have different access habits and behavior preferences,and there are large differences in access behavior for different functional points of the application platform.If resources are not allocated based on user behavior preferences,certain services will inevitably be caused,and some services will be overloaded,which will greatly affect user experience and system performance.This article takes the microservices architecture-based application platform as the background and conducts in-depth research on this issue:1.Propose a load prediction model based on user behavior preferences.Firstly,user historical access data and user private data are used to construct user portraits to learn user behavior preferences,so as to characterize user access routes at different function points.Finally,this model is used to predict the user's future access trends to function points.In addition,in order to use the information of service historical load,a hybrid forecasting model based on linear method and non-linear method is used for load forecasting.The complete load forecasting model combines the characteristics of user behavior preferences with a hybrid forecasting model based on service historical load.2.Propose a resource allocation model based on multi-agent reinforcement learning method.Treat each type of service as an agent,and introduce an attention mechanism based on the Actor-Critic algorithm for decentralized training strategies in a multi-agent environment.And combined with the load prediction model of the first part,the resource allocation model learns the best resource allocation strategy under different load situations.The final model can consume the least resources while meeting the needs of users.Finally,the paper evaluates the above methods through simulation experiments.Experiments show that the resource allocation method proposed in this paper has advantages over traditional resource allocation methods.Through the above research,this paper implements a new resource allocation method.An application based on the microservice architecture can design its own resource allocation method according to the method proposed in this article.In the case of meeting user needs,reduce resource consumption and avoid resource waste.