Research On Resource Allocation Algorithm Of Cloud Platform Based On Auto-scaling Technology

Today,most companies have begun to use cloud platforms to host their software applications.Cloud platforms are shared resources and can provide various services,such as infrastructure as a service(Iaa S).These cloud services are provided to users in the form of virtual machines(VM)to process users' own business.In the cloud platform,a physical machine can usually be divided into several VMs,which is very convenient for the management of VMs,and can be easily start and resize.Therefore,with the help of virtual technology,web applications can be dynamically deployed in the cloud environment,and web applications deployed on cloud platforms often face load changes.When faced with a sharp increase in load,more cloud resources need to be rented,while the usual load is lower.More resources will cause a waste of resources.The elastic function of the cloud platform enables the cloud platform to automatically scale the cloud resources it owns according to the load changes of the Web application,so that it can reduce costs while meeting its quality of service(Qo S).For this reason,various public cloud platforms that provide Iaa S services need to continuously upgrade and improve the elastic functions of cloud platforms in order to better serve customers and attract them to move software applications to the cloud.In order to realize the automatic scaling of cloud platform resources,this paper uses virtual machines as the resource allocation granularity for web applications deployed on the Iaa S public cloud to conduct an elastic scaling strategy research.The purpose is to ensure the Qo S of cloud web applications while reducing their use of resources as much as possible.In order to save costs.The main work of this paper is as follows:(1)Aiming at the slow convergence speed of existing elastic scaling algorithms based on reinforcement learning,the PDS-lambda elastic scaling algorithm is proposed.This algorithm comprehensively considers user service defaults,the current load of web applications and the number of virtual machines,and uses the post-decision state Divide the information that the algorithm needs to learn into dynamic known and dynamic unknown,so that the algorithm only learns the part of the dynamic unknown and uses multi-step updates to quickly achieve convergence,so that the algorithm can compare with existing elastic scaling algorithms based on reinforcement learning.Convergence is reached faster,thereby reducing costs.(2)In view of the problem that the workload of Web applications will change over time and it is difficult to achieve active automatic scaling,the AR-QT elastic scaling algorithm is proposed by combining the auto regressive model in the time series analysis method with the queuing theory.The algorithm can automatically predict the optimal number of virtual machines required by cloud web applications in a future period of time based on historical data,so that it can reduce operating costs as much as possible while meeting Qo S.(3)The two algorithms are simulated on cloudsim.The first experiment proves that the PDS-lambda algorithm can reach convergence faster than the reinforcement learning algorithm in related literature.The second experiment proves that the AR-QT algorithm can accurately predict and has better performance and flexibility than the related queuing theory algorithm.