Research On Container Cloud Resource Scheduling Algorithm Based On Kubernetes

Recent years,the rise of cloud computing has led enterprises to go to the "cloud".Meanwhile,microservices and container technology are widely used in the field of cloud computing.Kubernetes,currently the most mainstream container orchestration tool,is used by a number of cloud service providers as a solution.to application deployment.Container orchestration is its appearance,but its core function is resource scheduling.How to provide the resources in the cloud platform to users in an optimal way is an urgent problem to be solved.A good resource scheduling strategy will save enterprises a lot of costs.This article will take the Kubernetes Scheduler as the research object,and propose corresponding optimizations for the short comings of the existing scheduling algorithms of the scheduler.The main work content is as follows:(1)The resource scheduling algorithm defaulted by Kubernetes determines the score of the node at the optimization stage only as per the node's CPU and the utilization rate of memory,namely the resource utilization rate of single node is considered.The excellent nodes in cluster will be dispatched more Pods so that the quantity of Pod at other nodes is much less and the load balance of cluster resource can't be guaranteed.In response to this problem,a Kubernetes resource scheduling algorithm that is based on improved genetic algorithm(IGA)is proposed.The algorithm adds two evaluation indicators of network bandwidth and disk IO.Different weight values are assigned to the evaluation indicators.On the basis of the standard genetic algorithm,the calibration dictionary is introduced for calibration,and the individual failing to comply with the user configuration in the stage of randomly initializing the population,cross stage and variation stage is repaired.The results of experiment illustrated that in the comparison with t-he Kubernetes default resource scheduling algorithm,this algorithm takes the resource utilization of all nodes into consideration in the cluster,and has a better effect in ensuring cluster load balancing.(2)As for the current Kubernetes scheduling(a static scheduling mechanism),four dynamic resource scheduling trigger mechanisms are put forward: timed trigger,regular trigger,node capacity expansion/capacity shrinkage trigger and node load super threshold trigger.Based on scheduling of modified genetic algorithm,the cluster load balancing can be still ensured when the node resource load changes by the dynamic trigger mechanism.(3)Because plenty of computing resources of the Master node are occupied when the modified genetic algorithm runs,the normal operation of other components on the major node is influenced.As for this problem,the distributed resource scheduler based on RPC is proposed to transfer out the computing task from the major node on the basis of not changing the original scheduling mechanism.(4)As for the modified genetic algorithm running slowly in the uniprocessor,the parallelization of the improved genetic algorithm on the basis of Spark(SP-IGA)is proposed.Through the analysis of Spark's parallel mechanism,the parallelization of the improved genetic algorithm which is proposed in this thesis is realized based on the two computing parallel and data parallel modes,and the running speed of the algorithm has been improved.This thesis contains 43 figures,7 frames,and 64 references.