Application Research On The Estimation Of Distribution Algorithm For Scheduling Cloud Computing Resource

With the rapid development of information technology and the growing demand for computing and storage in big data environments,cloud computing which is a novel mode of resource utilization enables end-users to easily purchase convenient services on demand in the pay-per-use service model.Due to cloud provides a finite pool of virtual-ized computing resources,how to "optimally" schedule cloud computing resources has become a significant and valuable topic in scheduling field,which needs to satisfy the various requirements of both users and cloud providers as far as possible.At present,the existing related algorithms for solving the similar problems are mainly based on evolu-tionary computation approaches and can not consider the potential inter-dependencies between cloud tasks during scheduling.However,compared with traditional intelligent computation methods,the estimation of distributed algorithm can learn probabilistic model from dominant populations,and has incomparable advantages in constructing the variable-dependent relationship model.Considering that the cloud computing re-source scheduling problem has a strong practical application background,designing a reasonable and efficient resource scheduling algorithm has important practical signifi-cance for improving user satisfaction and supplier benefits.This paper starts with the optimization problems of QoS on the SaaS layer,and studies the cloud computing resource scheduling problems considering specific opti-mization objectives and constraints,and then designs the corresponding effective esti-mation of distributed algorithms according to the character of the investigated problems.Firstly,this article give a systematic instruction of the problem background and related works.Next,the classification of cloud computing resource scheduling problems is in-troduced briefly and the frequent algorithms for solving those problems are described.Finally,the two specific problems investigated and the corresponding estimation of dis-tributed algorithms are presented in two chapters in detail.To minimize makespan for scheduling independent tasks in cloud computing,an estimation of distributed algorithm without dependence is improved after problem for-mulation,Our improvements mainly include the following aspects:encoding scheme,learning rate,probabilistic model,initialling population and sampling strategy,more-over,an improved genetic algorithm is applied.The experiment results show that the proposed algorithm not only gets better solution,but also has faster convergence speed.Next,we aim at minimizing the user costs by regarding the deadline as a constraint for scheduling independent tasks,to tackle the investigated problem,an effective and variable-dependent hybrid estimation of distributed algorithm based on Markov chain model is proposed.In this algorithm,the concept of virtual machine selection diver-sity is innovatively proposed,and two different methods are considered to solve the marginal probabilities.In addition,a real-time heuristic information and the amelio-rated strategy for probabilities calculation are both applied to construct the conditional probabilities.At last,a simple tabu search algorithm is employed to keep the proposed algorithm maintaining a higher diversity at the sampling stage.The experiment results show that the proposed algorithm not only obtains the best solution quality but also has competitive convergence and makespan among all compared algorithms.Finally,the research results of this paper are summarized and the prospects for further research directions are presented.