Container Cloud Elastic Scaling Strategy Based On Dynamic Thresholds

Cloud computing is a kind of shared resource pool that connects servers together to form on-demand allocation and convenient access,so that it can not only obtain low cost but also have the performance of a supercomputer,providing basic support for the development of big data,artificial intelligence,Internet of Things and other fields.Three layers in the traditional cloud computing service mode,the resource scheduling unit of Iaa S layer is a virtual machine,so the existing scheduling speed slow,software stack problem of environment is not the same and low resource utilization,and because of its traditional Paa S platform is based on the Iaa S layer,so the program environment and architecture and so on is limited by the defects of the virtual machine itself is very big.Therefore cloud computing urgently needs a new,efficient and convenient virtualization technology to solve the above problems.In recent years,with the rapidly development of Docker container technology,container-based virtualization technology has become the first choice of various cloud computing manufacturers.Among them,how to allocate the resources of container cluster reasonably and effectively is the key to improve user service level and also the focus of cloud computing research.At present,in the container cloud domain,it is difficult to rapidly expand capacity according to the load of container cloud in the face of sudden heavy traffic and delay capacity reduction as much as possible when the load of container cloud decreases.Such problems will greatly improve the SLA default rate of users and reduce the service quality of container cloud.Elastic expansion and contraction of container cloud is a key part of container cloud.It can expand and shrink in time according to the load of container cloud so as to achieve load balancing and ensure the stable operation of container-based micro-services under different load conditions.Container cloud elastic scaling can be divided into horizontal elastic scaling and vertical elastic scaling.The horizontal elastic scaling increases or decreases the number of containers in the container cloud to achieve capacity expansion,while the vertical elastic scaling improves the performance of a single container to achieve capacity expansion.However,capacity expansion in the container cloud has a certain response delay.In the case of rapid traffic changes,delayed capacity expansion may result in SLA defaults of a large number of cloud tasks,reducing the service quality of the container cloud.Therefore,a new container cloud elastic scaling strategy can be proposed by monitoring container cloud load changes,which helps reduce SLA default rate of users and improve service quality of container cloud.In order to solve this problem,this paper proposes two capacity expansion and shrinkage optimization algorithms for container cloud by combining threshold-based responsive elastic scaling strategy and prediction-based elastic scaling strategy.(1)The horizontal scaling algorithm of Kubernetes(Horizontal Pod Autoscaler,HPA),the mainstream container cloud orchestration tool,cannot expand and shrink in time in the face of sudden large traffic,resulting in increased SLA default rate of users and decreased service quality of container cloud.This paper proposes a horizontal elastic scaling strategy for container cloud based on dynamic threshold prediction(Dynamic Threshold-Horizontal Pod Autoscaler,DT-HPA).The second moving average method is used to predict the resource demand of the cluster in the future time according to the load change rate of the container,and the threshold difference between the dynamic threshold and the static threshold is called the fault tolerance zone,and the error of prediction can be reduced through the fault tolerance zone.At the same time,during the prediction interval,to prevent the cluster load from being too high due to burst traffic and thus increasing SLA default rate,the container capacity is expanded and shrunk based on the dynamic threshold.Simulation experiments show that compared with DT-HPA algorithm,the DT-HPA algorithm proposed in this paper can achieve rapid capacity expansion and delayed capacity reduction and effectively reduce the SLA default rate.(2)Kubernetes' s Vertical elastic scaling algorithm(Vertical Pod Autoscaler,VPA)adopts a fixed expansion and contraction mechanism to carry out elastic scaling,This mechanism can no longer meet the vertical scaling requirements of containers in today's environment of increasingly complex user requests.This paper proposes a vertical elastic scaling strategy for container cloud based on dynamic threshold prediction(Dynamic Threshold-Vertical Pod Autoscaler,DT-VPA).When the container load changes rapidly,the historical load of the container is counted and the resource demand of the cluster in the future is predicted by using the quadratic moving average method combined with the available resources of the current server,and the prediction error is reduced by using the dynamic threshold.Simulation experiments show that compared with DT-VPA algorithm,the DT-VPA algorithm proposed in this paper can achieve rapid capacity expansion and delayed capacity reduction and effectively reduce the SLA default rate.