Research On Scheduling Algorithm For Optimizing Cost-efficiency In Geo-Distributed Cloud Systems

More data centers distributed across regions recent years.The cost of data centers construction varies in different regions for the diversity in housing prices,wage levels and electricity charges.Therefore,cloud providers offer different pricing to users according to their sites.Different pricing of cloud services brings opportunities to users that they can schedule their tasks to the appropriate data centers to reduce their cost.The reduction of cost will drive more and more users to apply cloud services,and that would promote the prosperity and development of cloud computing.In this paper,we examine the problem of minimizing cost for cloud users with respect to geo-distributed cloud systems.By modeling the problem as a general assignment problem(GAP),we apply the Augmented Lagrangian Multiplier Method(ALMM)to obtain a schedule solution.However,ALMM does not converge to a feasible solution in cases,we therefore devise an Adjusting algorithm which adjusts the results produced by ALMM to make it feasible.Moreover,as the convergence speed of ALMM is relatively low,we further devise a Decreased Value Density Scheduling algorithm(DVDS),which is able to obtain a result schedule in a rather short time.In order to make a better comparison,this paper compares the general greedy strategy with the DVDS algorithm,and the experiment shows that the general greedy strategy is more expensive than the DVDS algorithm,about 6%~11%.We apply our algorithms to both linear pricing model and the more realistic piecewise pricing model,and find that piecewise pricing model complicates the problem by suggesting more combinations of virtual machine(VM)and bandwidth configurations.Experimental results show that,DVDS algorithm produces higher cost than ALMM algorithm,but it saves more time than ALMM algorithm.When the number of tasks increases,the efficiency of the DVDS algorithm is significantly faster than that of the ALMM + Adjustment algorithm,which is only about 1% of the ALMM + Adjustment algorithm.