Research On Elastic Scaling Based On Load Forecasting In Container Cloud Environment

Nowadays,the widespread application of container cloud makes application software development can be practiced with the idea of Dev Ops.In this process,whether the resources in the container cloud environment can be used efficiently and whether the microservice applications deployed in the cluster can guarantee Qo S becomes particularly important.In actual production,in the face of changes in the load in the container cloud environment,threshold-based reactive scaling strategies are often used to dynamically adjust the cluster.However,traditional reactive scaling strategies cannot be implemented in a timely manner when the load is too large.The response time of the request is too long.When the load is too small,the rapid shrinkage of resources cannot be achieved,which will also cause waste of resources.Therefore,this paper uses predictive elastic scaling to perform resource scheduling in a container cloud environment.The detailed research work of this paper is as follows:(1)A load prediction model based on ensemble learning is proposed.First,a dynamic load model is constructed for different microservice applications in the container cloud environment.Then,based on the characteristics of the linear and non-linearity of the load,the ARIMA load prediction model was used to predict the linear part of the load.At the same time,the lightning attachment optimization algorithm was improved to construct a dynamic LSTM prediction model to predict the non-linear portion of the load.The weight method combines the weighted results of the two parts,and finally improves the accuracy of the prediction model by error-improving the combined prediction results.The proposed load forecasting model provides a basis for the study of elastic scaling and resource limiting strategies.(2)Propose a container cloud elastic scaling and resource limitation scheme based on load prediction.First,the load forecasting model proposed in this paper is used for load forecasting.Then,based on the obtained prediction results and the current load situation of the cluster,resource limits and elastic scaling are performed.Aiming at the problem that the surge in load may cause OOM,by studying the node's future load and current node resource usage,the available resources of a single microservice container are limited to prevent the microservice container from being killed by the kernel process because it consumes too much resources.At the same time,by analyzing the future load and current load of each service in the cluster,the microservice container in the cluster is scaled to improve the resource utilization rate and service quality of the cluster.Conducting relevant experiments From the results,it can be seen that the experiments proposed in this paper significantly reduce the request response time and improve the utilization rate of container cluster resources.(3)A microservices elastic scaling system based on Dev Ops is constructed.Based on the load forecasting model and elastic scaling and resource limit scheduling algorithm proposed in this paper,a microservices container cluster scheduling system is constructed on the basis of the Docker Swarm container cluster management platform.Realize real-time monitoring of node and container resources in the cluster,and at the same time use existing service management tools to achieve load balancing of container clusters for service discovery and service registration issues brought by elastic scaling.At the same time,the management of container images and the continuous integration and continuous release of services have been strengthened.