Research And Implementation Of Kubernetes Scheduling Algorithm

With the development of cloud computing and container technology,the docker as the core container technology is rapidly becoming widely adopted among developers and technology companies.Kubernetes has become the de-facto container cluster management system with its rich enterprise and production-level features.However,Kubernetes' default scheduling algorithm mainly has two main problems.One is that it does not take network IO into account,which is easy to cause problems such as unbalanced network IO of cluster nodes and network IO performance bottleneck of individual nodes.Another is that Kubernetes does not consider multi-criteria resource such as disk IO,network IO and bandwidth,which easily causes the problem of unbalanced multi-criteria resource allocation of the cluster.To address these problems,this paper makes the following con-tributions.By studying Kubernetes scheduler architecture and mechanism,this paper proposes and develops a scheduler extender named Liang based on Kubernetes Scheduler Extender mechanism.Then we implement network IO balanced scheduling algorithm and multicriteria resource scheduling algorithm based Liang.Balanced Net IO Priority(BNP)adds network IO resource request and combines the network information of candidate nodes to select the best node.BNP makes the overall network IO usage of the cluster more balanced and reduces the container deployment time.Multicriteria resources scheduling algorithm CMDN is based on TOPSIS decision algorithm.The CMDN algorithm takes CPU utilization,memory utilization,disk IO,network IO and NIC bandwidth of candidate nodes into account.It then scores nodes comprehensively using TOPSIS algorithm which brings two scheduling effects of multi-criteria resource balancing and compactness.The experiments show that BNP Algorithm improves the balance level of cluster network IO,prevents nodes from network IO bottlenecks,and also reduces the container deployment time by 32%.The CMDN Algorithm can balance the multi-criteria resource utilization such as CPU,Memory,disk IO and network IO of the cluster nodes in balanc-ing policy.It also reduces container deployment time by 21%.The CMDN Algorithm can schedule containers to the nodes with high multidi-criteria resource utilization in the compact policy which achieves the expected results.