Research On Modeling And Algorithm Design Of The Multi-object Resource Sheduling Problem In Cloud

Under the cloud computing system,cloud users can submit tasks on demand,while the service providers would process physical resources into abstract resource pools,which undertake tasks.However,with the scale of tasks increasing and the complexity of cloud architecture evolving,some performance problems and energy consumption problems appear in the datacenter,and it is difficult to meet the diverse requirements.Cloud task scheduling and resource allocation scheme directly affect the performance of the whole system,considering the cloud model complexity and variety of problems.In this article,scheduling problem in cloud will firstly be divided into task-virtual nodes level and virtual node-physical machine layer,and the study of modeling and algorithm,considering different scheduling goals.At the task-to-virtual computing node scheduling level,the relevant cloud taskresource model is established,considering tasks execution time,user execution cost and the overall load of the system.Besides,a multi-objective cloud task scheduling algorithm based on ant colony algorithm is proposed.To avoid blindness in initial search phrase,prior information about the nodes is added in the pheromone initialization stage and the heuristic min-min algorithm is combined to improve the quality of initial solution.At the beginning of the search,a global pheromone update strategy with elite method is used to strengthen of pheromones to speed up the search efficiency,and a kind of local pheromone update strategy which has the function of load regulation is put forward,aiming at adjusting pheromone level influence subsequent task node distribution tendency,to make the system load balanced.In view of the disadvantage that ant colony algorithm is prone to fall into the local optimum due to the positive feedback mechanism in the late stage,the adaptive volatile factor is used to suppress the positive feedback characteristics in the late stage of decision search,so as to improve the global searching ability to search the solution space more fully.Experimental results show that compared with RR,FCFS and PBACO,the algorithm designed in this paper has advantages in total task execution time,user cost and system load balancing degree.At the scheduling level between the virtual computing node to and the physical host,energy consumption and service default level are considered as the scheduling goals.a related virtual node-energy consumption model is established and a multi-objective virtual machine integration algorithm based on the historical load information of virtual nodes and the overall state of cluster is proposed.Considering the historical load information,the load trend of virtual nodes on the overload physical machine is estimated by Cox-Stuart test and ADF trend test,and the virtual nodes on overload physical machines are divided into the sets with different priorities.On the basis of the priority of the collection,considering the error of trend judgment and load fluctuation,a suitable migration target selection method is proposed for the nodes of different collections.In addition,considering on that the selection of the last virtual machine to be migrated has a great impact on the state of the physical machine after the migration,the benefit assessment is carried out.Moreover,the load of data center is often concentrated in the later stage of integration,as a result,a second judgment based on cluster congestion degree is proposed to avoid excessive migration at this time.Finally,the experimental results show that the proposed virtual machine integration algorithm can reduce energy consumption and service violation more effectively.