Research And Implementation Of Workload Prediction And Auto-scaling In Container Cloud Platform

With the rapid development of cloud computing,more and more companies use microservice architecture to develop business systems and deploy them on cloud platforms.The microservice architecture splits the complex system into multiple independent subsystems,reducing the degree of coupling between different functions of the system.Container technology packs microservices and their dependent environments into mirrors,shares mirrors through mirror warehouses and downloads and runs them,reducing the cost of microservices and improving flexibility.Due to the continuous increase in the number of containers,the container cloud platform came into being.The container is scheduled and managed in the cluster,and microservice instances are created or deleted in the cluster through automatic scaling technology to realize the allocation of resources on demand.However,the current automatic scaling strategies adopted by the container cloud platform all belong to responsive scaling,which has a lag,leads to an increase in the response time of microservices and waste of resources.Therefore,this thesis proposes a microservice workload prediction model,designs and implements a predictive automatic scaling scheme.The specific work is as follows:(1)The design of microservice workload prediction model.This model predicts the load of microservices through Sequence Encoding,Sequence Feature Extraction and Sequence Generating,referred to as SEFEG prediction model.Compared with the existing workload prediction model,the SEFEG model predicts that the performance after 5 minutes,10 minutes,15 minutes and 20 minutes is better than the other three comparison models.(2)The design of predictive auto-scaling scheme for microservices.In view of the shortcomings of the current container cloud platform scaling strategy,a predictive auto-scaling architecture is designed,and the instance discovery module,load index collection module,workload prediction module and auto-scaling module in the architecture are designed in detail.The solution can be based on the predictive model Realize auto-scaling.(3)The realization of predictive auto-scaling scheme based on Kubernetes.Using Custom Resource Define(CRD)and Custom Controller,a predictive auto-scaling solution is implemented,including an instance discovery module,an indicator collection module,a prediction module,and an auto-scaling module.The results of the steady load test and the burst load test show that the implemented solution is better than the built-in auto-scaling solution of Kubernetes,which reduces the response time of microservices.