Multi-Objective Job Scheduling For Cloud Computer Heterogeneous Resources

With the development of the cloud computing,user's demand for cloud computing also change rapidly.At the same time,variety types of resources are introduced by the expanding of cloud computing platform.In cloud computing platform with large number of heterogeneous resources,job scheduling is NP hard problem.In the scheduling process,job scheduling with multi-objective constraint is more difficult,considering the execution time,power consumption,cost,SLA and many other constraints.In this thesis,we discuss the two different ways of user submission job,which can be divided into static and dynamic submitting.After definition the scheduling objectives and constraints of two submitting ways,this thesis designs the scheduling algorithm,and experiments in the simulation platform.The main work of this thesis is as follows:(1)We design a static submission job scheduling algorithm based on accelerating convergent bee colony for the heterogeneous resource platform of cloud computing.First,we establish the centralized model of static job submission under the heterogeneous resources environment.Also we define the multiple objectives and constraints,such as execution time,energy consumption and cost.By using the advantages of colony algorithm parameter settings and good performance,we design the scheduling algorithm based on the accelerated convergence bee colony.By constructing the multi-objective benefits nectar matrix,the heterogeneous resources load balancing strategy is introduced in the initializing nectar and exchanging phase.Moreover,we use chaos search and tabu strategy to accelerate the colony,reduce the scheduling time.Comparing with other scheduling algorithms such as ant colony,particle swarm,tabu search,the performance test on CloudSim show that the proposed scheduling algorithm based on accelerated convergence bee colony have fast convergence speed and better stability.(2)We design a dynamic submission job scheduling algorithm based on Q-learning for the heterogeneous resource platform of cloud computing.We build the queuing model based on interval time and number distribution.After setting the waiting time,energy cost and other scheduling objectives and SLA constraint of dynamic submit,Markoff decision process four tuples is utilized for analysis the scheduling objectives,and determine the reward size for each scheduling.The reinforcement learning scheduling algorithm based on Q-learning utilizes the discount reward cumulant of the prior scheduling scheme,for scheduling the subsequent arrival job.(3)We compare the capabilities of different scheduling algorithms in static and dynamic submission job scene.We build a cloud computing platform with 40 virtual machines by heterogeneous resources simulation platform MultiRECloudSim,which is the extended platform of CloudSim.Then,carrying on the comparison test for the accelerated convergence bee colony scheduling algorithm and other scheduling methods,compared with the tabu search can reduce execution time by 8.63%,control the platform energy consumption and cost.Also,we use the event handling mechanism,setting up a job dynamic arrival flow in the simulation platform.After that,we contrast test the Q-learning scheduling algorithm and greedy,fair scheduling algorithm.The simulation results show that the algorithm can meet the SLA constraints,compared to the greedy 10.36% fewer job wait time and the execution time of 5.59%.