Makespan And Cost Optimization In Scientific Workflow Scheduling In Cloud Computing Environment

As a new commercial model and computing model,cloud computing promotes the progress and development of society.Cloud computing provides computational resources,storage resources and software as service to the user through virtualization technology.Users no longer need to purchase the underlying hardware;they can get the resources from service provider on demand.Scientific workflow is used to manage scientific research applications,due to its massive computing and data storage needs,scientific workflow is often executed on distribute compute system.As the ease and low cost of cloud computing,scientific workflow is take advantage of cloud computing platforms.The execution of scientific workflow usually needed to meet the requirement of users,so optimizing the schedule of scientific workflow is a critical research problem.The thesis first introduces cloud computing and scientific workflow,and then conducts a comprehensive analysis of existing schedule algorithm.In order to overcome the disadvantages of existing works,the thesis do the following works:(1)First,we proposed to use multi-objective evolutionary algorithms NSGA-II hybrid with local search to solve the multi-objective scientific workflow scheduling in cloud computing environment.Usually user wills specific their needs when executing the scientific workflow,make-span and economic cost are two common goals of scheduling,and yet these two objectives conflict with each other.Previous works try to optimization one objective while qualify another objective or combine two objectives into one objective using weighted sum method,and they only provide one schedule solution to the problem.The thesis model the scientific workflow as a dependency graph,and provide multiple Pareto optimal schedule solution using NSGA-II algorithm,so user and select suitable schedule solution according to their needs.We make two improvements on the scheduling algorithm.Due to the massive tasks and short tasks execution time in scientific workflow and the hourly billing mode of cloud,we present a task aggregation algorithm to optimize the workflow make-span and economic-cost.A new local search algorithm is proposed to further explore the schedule solution in solution space.(2)Secondly,in the experimental validation phase,we use CloudSim as a cloud computing simulation platform,and make some modification to incorporate scientific workflow scheduling.We compare the proposed algorithm with other algorithms under no-constrained condition and constrained condition,experiment shows that the proposed algorithm NSGA-? is able to obtain more and better Pareto optimal solution and show effectively under tight constrained condition.Task aggregate simulation experiment shows that make-span is reduced by about 18.4%and economic cost is reduced by about 3.4%.The execution time of getting the Pareto optimal solution is 27.5%less.When NSGA-II is hybrid with the local search algorithm,it tends to get better Pareto optimal solution.