Research On Key Technologies Of Container Scheduling For Computer Cluster

A computer cluster is a complex computer system that connects a set of computer software and hardware through the network to work together.At present,container technology has brought great breakthroughs to the performance improvement of computer clusters,and major Internet manufacturers are stepping up the development of new computer clusters based on container technology.With the continuous promotion of computer clusters,some defects of computer clusters have become more and more prominent,such as container scheduling lag,unbalanced load between nodes,complex data exchange links between containers,and slow service response,which greatly limits their development.In view of the current problems in computer clusters,this study carries out the research on load prediction algorithm and container scheduling algorithm of computer clusters,and the main contents of this study include the following aspects:(1)Aiming at the problem of container scheduling lag during the operation of computer clusters,a load prediction algorithm of extreme learning machine based on conditional entropy and genetic algorithm is proposed.The load data of each node is collected into the time series database,the numerical sequence is granulated by fuzzy Cmeans clustering,the model order is determined by conditional entropy after obtaining the category sequence,the hidden layer neuron of the extreme learning machine is optimized by the genetic algorithm,and the load prediction is completed by the improved extreme learning machine,the prediction accuracy is higher than that of similar algorithms.(2)Aiming at the problem of unbalanced load between nodes,complex data exchange links between containers and slow service response during the operation of computer clusters,this paper mathematically models nodes.An undirected weighted graph is used to represent the data exchange relationship between containers,vectors are used to represent the resource requirements of containers and the resource capacity of nodes,the definitions of load imbalance and cross-node data exchange rate are given,and the objective function and constraints of container scheduling are given.The swarm intelligence algorithm is used to solve the scheduling scheme for reducing the cross-node data exchange rate and load imbalance.(3)Aiming at the problem that the traditional particle swarm algorithm cannot balance local search and global search,and is easy to fall into local optimality,a multistrategy particle swarm algorithm(MSPSO)is proposed by using strategies such as mutual information and Levy perturbation,adaptive weighting and adaptive learning factor.Experiments show that the search ability of MSPSO is superior to that of traditional particle swarm optimization and some of its improved algorithms.Finally,the MSPSO and entropy weight method are combined to solve the container scheduling scheme.Finally,an experimental platform is built based on the existing equipment and open source software of the laboratory,and the feasibility of the load prediction algorithm and container scheduling algorithm proposed in this paper is verified on this platform.