Research And Implementation Of Elastic Scaling Scheme For Combinatorial Predictive Containers

With the advancement of cloud computing technology and the rapid development of container technology,the concept of micro-service architecture has gradually become popular in the industry,and more and more micro-service applications are deployed to container environment.The main purpose of micro-services is to reduce the coupling between functions by breaking down a large functional module into multiple services that can be run separately.Using container technology to package the micro-service and its operating environment uniformly can reduce the operation cost and resource cost of the platform,but it also brings new challenges to the management of the container platform.As the size of the container expands,the monitoring object of the platform changes from a single service application to a container and a service application running in the container,which complicates the monitoring management of the container platform.Moreover,since the services running in the containers vary and the demand for resources is inconsistent,how to allocate resources properly will be a major challenge that needs to be solved urgently.In addition,there will be another problem about how to manage this log datum uniformly because of the diverse formats of log data generated by each service and the exponential growth in the amount of data.In view of this series of problems,the author makes an in-depth study from the following three aspects: firstly,on the basis of Spring MVC architecture,the author uses Prometheus technology,cAdvisor technology,Flume technology to realize the performance monitoring of container platform and the unified management function of log data.Secondly,aiming at the problem of rational allocation of resources in container platform,the container elastic scaling scheme based on combinatorial prediction model is proposed,by in-depth study of the existing responsive elastic scaling scheme and predictive elastic scaling scheme.Combining the advantages of ARIMA model and SVM model in short time series prediction,a combined predictive algorithm model based on ARIMA and SVM is designed.On the basis of monitoring service,the performance data of container is used as the data source for multiple training experiments.Compared with the existing predictive algorithm model,the accuracy of combinatorial prediction algorithm is improved by nearly 10%,and its accuracy is up to 91.95%,which provides a powerful judgment condition for realizing elastic scaling of predictive container.Thirdly,combined with the data source provided by monitoring service and the trained combination prediction algorithm model,the predictive elastic scaling function is designed and implemented.By deploying monitoring services,log services,and elastic scaling services to a container environment for testing,the results show that the combined predictive elastic scaling scheme proposed in this thesis is more excellent in the rational utilization of resources than the responsive elastic scaling scheme and the predictive elastic scaling scheme based on ARIMA model and SVM model.What's more,the monitoring service and log service designed in this thesis also solve the problem that the current container platform is not easy to monitor and information is scattered.And it is fully proved that the theoretical significance and practical value of this research.