Design And Implementation Of Scheduling Strategy For Distributed Resource Scheduling Platform

Container is a virtualization technology emerging in recent years,which is widely used in the field of cloud computing.It greatly facilitates the deployment of cloud computing applications.More and more applications turn to the use of container development and deployment.Among many container choreography management tools,kubernetes is becoming a de facto standard.It provides a set of cluster self recovery mechanism,including automatic restart,automatic scheduling and automatic backup of containers.However,kubernetes' default scheduler relies on the resource allocation requirements provided by the application owner,and does not consider the real-time resource usage.And with the application of multi cluster architecture,it is usually necessary to run multiple kubernetes clusters,which increases the complexity of management.After deeply studying kubernetes and docker container technology,this paper summarizes the shortcomings of kubernetes in scheduling methods and strategies,and designs and implements a distributed resource scheduling management platform based on actual production application requirements,which decouples services from specific machines,provides resource isolation and automatic resource allocation between services,and makes it easy for service mixing.The system can select the appropriate host deployment according to the real-time resource usage of the host in the resource pool.This paper elaborates the theoretical basis,ideas and implementation scheme involved.The main work of this paper is as follows.1.This paper focuses on the study of resource scheduling technology and the design and architecture of mainstream cluster scheduling system,and studies and summarizes many container scheduling management tools on the market.It is found that kubernetes,as a representative of container management platform,does not consider the use of real-time resources when scheduling,and scheduling is mainly oriented to single cluster.Due to the limitation of CGroup mechanism of docker container technology,This paper analyzes the advantages and disadvantages of cloud computing,and proposes the demand of off-line hybrid scheduling,which includes resource over dispatch,multi-dimensional resources and multi priority preemption.2.Complete the realization of the main functions of the system combined with demand analysis.A multi cluster oriented two-tier scheduler is implemented,which can preempt resources according to priority and do partial over dispatch.It normalizes heterogeneous quotas such as CPU and designs ng coefficient for heterogeneous quota scheduling.