Research On Serverless Architecture Scheduling Strategy

As a new type of cloud computing paradigm,the serverless architecture further simplifies the cloud computing model and represents another revolution in the field of cloud computing.The serverless architecture is essentially different from the traditional cloud computing paradigm.In the serverless architecture,computing and storage are decoupled from each other,and they scale and operate independently.At the same time,the executed code is used as the running object,and the user does not need to care about the underlying resources required for the code to run,and the underlying platform is responsible for processing.This approach improves the developer's productivity and reduces operation and maintenance costs.Knative is a serverless framework build on the kubernetes platform,used to build,deploy and manage serverless applications.The applications hosting service provided by the knative framework can reduce the complexity and risk of directly operating kubernetes clusters and improve the efficiency of application iteration and service delivery.Knative framework completes the serverless application scheduling mainly divide into two stages.In the first phase,the knative pod autoscaler(KPA)module calculates the number of Pods that need to be started based on the number of real-time task requests.In the second phase,the kubernetes cluster completes the task scheduling based on the number of pods calculated by the KPA module.At present,the knative framework is very perfect for the scheduling process of serverless applications,but there are still areas that need to be improved and optimized,mainly in tow aspects:(1)This paper deals with past access requests based on double exponential smoothing,optimizes the algorithm for calculating the number of Pods in the KAP module,and reconstructs the KPA module based on the optimized algorithm.The comparison experiment shows the MAE(Mean Absolute Error)and the MSE(Mean Squared Error)between the number of Pods calculated by the optimized KPA module and the number of ideal Pods are significantly reduced.(2)In response to the redundant computing problems existing in kubernetes default scheduler,this paper redesigns the kubernetes scheduling process to build pods with the same priority and resource requirements and limitations into a Pod group,with the Pod group as a basic scheduling unit for scheduling.In each Podgroup scheduling process,the group can make full use of the calculation generated during the first Pod scheduling in the group,reduce unnecessary redundancy calculation,reduce the wast of computing resources.Experiments show that the kubernetes scheduler,which is rebuilt based on the optimization design scheme,effectively reduces the redundant calculation in the scheduling process and improves the efficiency of the Pod scheduling.Through the above two aspects,we can optimize the problems in the knative framework scheduling process,and further improve the efficiency of the framework task scheduling.