Design And Implementation Of Function Launching Accelerator In Serverless Cloud Computing Platform

Serverless computing has emerged as a new compelling paradigm for the deployment of applications and services.In serverless computing,the application is split into multiple fine-grained functions.Users only need to focus on their business logic and do not need to worry about operation and maintenance issues such as server configuration,maintenance,update and expansion.The serverless cloud computing platform adopts an on-demand structure,which means that the platform loads functions only when service requests or events are triggered,and the idle function instances are deleted at runtime.The on-demand structure can reduce the platform resource load,but serious delay can occur when the function is invoked again after the function container is deleted.This phenomenon is called cold start.Cold start severely affects the responsiveness of serverless computing platform and limits the use and adoption of serverless computing by a broader range of applications.This paper focuses on serverless computing platforms and the function launching acceleration strategies in the platform.The purpose is to build a mature serverless cloud computing platform and implement a function launching accelerator to solve the cold start problem.Firstly,this paper studies the architecture,service methods and workflow of serverless cloud computing platform,and we design and implement a serverless cloud computing platform based on Kubeless.The platform combines resource management capability of Kubernetes with the functional computing capability of serverless framework.The platfonn implements the automatic deployment and operation of function codes,and provides users with basic functions of resource management.Secondly,in order to solve the cold start problem in serverless cloud computing platform,this paper designs and implements a function launching accelerator.The accelerator uses Long Short-Term Memory networks and regression method to analyze the function invoking time series,and thereby preheats the functions to reduce launching delays.In addition,we introduce an adaptive container pool strategy to prevent preheating failures.This accelerator comprehensively considers the launching optimization effect and resource consumption,thereby improving the overall performance of the platform.Finally,we conduct extensive experiments to evaluate our strategy.The evaluation results demonstrate the effectiveness of our strategies.