The Optimization Of Scheduling Algorithm With Resource Affinity Based On The Docker Swarm Cluster

With the popularity of cloud computing,distributed computing technology has become enterprise customer the first choice of the background in the era of big data technology solutions,Internet companies have been distributed computing technology to daily business systems.However,as is known to all,cloud computing has a classic three layer model,IaaS(Infrastructure as a Service),PaaS(PlatForm as a Service),SaaS(Software as a Service),line is clear,clear responsibilities.But with the development of science and technology,relying on lightweight virtualization technology Docker has been up in the field of cloud computing,it blurs the boundaries between IaaS and PaaS,breaking the original classification of blunt,partly overturned classic cloud computing architectures,for cloud computing services in the form of development has brought limitless possibilities.the cluster management tool named Docker Swarm is duly became simple and efficient application delivery platform and cluster resource management solutions of today's Internet.On this basis,a set of complete and reliable orchestration engine algorithm which can give full play to the advantages of cluster is very important.However,for the moment,the choreographer engine behind the Docker(SwarmKit)included with the scheduling algorithm in heterogeneous resources is not ideal,can't perfectly realize the load balance,is not very good in line with expectations in terms of resource utilization and the clustering performance.Currently,Docker Swarm has three basic scheduling policies(spread,binpack,and random),each of which executes a container with a fixed number of resources.A dynamic weighted scheduling algorithm is proposed to solve the problem of cluster load balancing and cluster utilization in this paper.The algorithm weights the resources.Use parameter "bias" to dynamically adjust weights for different services.Calculate the weight of the node according to the resource utilization of each node.This value represents the load of the node and is used for scheduling.The experiment is compared,and the proposed algorithm has no parameter adjustment with the original scheduling strategy and weighted scheduling strategy.The results show that the algorithm can better balance the resource utilization of the nodes in the cluster.In addition,when the cluster is under high load pressure,using the proposed algorithm,the service runs faster.