Research On Multi-objective Workflow Scheduling Method In IaaS Environment

With the rapid development of science,infrastructure as a service(IaaS)is becoming the most important service model of the Cloud computing.It is the promising computing platform for scientific workflow computations.Cloud service providers can provide large-scale virtual computing resources to users,while at the same time,they might face resource scheduling problem with minimized cost to complete the workflow execution.In this dissertation,we study the multi-objective workflow scheduling in IaaS environments.The main contributions of our work include:(1)A multi-objective algorithm MOHEFT*based on MOHEFT is presented.The algorithm classifies the virtual machines waiting for scheduling according to their types,which significantly reduces the number of virtual machines.Moreover,it saves a large number of iterative operations and also reduces the time complexity for the algorithm.Therefore,based on the real-world applications published by Pegasus project with virtual machine configuration from Amazon EC2,and the experimental results prove the efficiency of our algorithm.(2)The multi-objective algorithm based on decomposition(MOSC/D)which optimizes the makespan and cost simultaneously is proposed.This algorithm combines the design of both list-based heuristic algorithm and multi-objective evolutionary algorithm.In addition,using a decomposition method,it decomposes the multi-objective optimization problem into a set of single objective optimization problems.However,this algorithm focuses on solving the single-objective problems which makes the scheduling process more effective.Therefore,compared to the MOHEFT and the NSGA-?*algorithms,our experimental results show that our proposed MOSC/D algorithm achieves better Pareto fronts with lower time complexity.(3)Lastly,to solve the time and budget constrained workflow scheduling problem in IaaS,we proposed a new efficient genetic algorithm named TBAI algorithm.We design basic genetic operations including the encoding,fitness evaluation and population initialization.Furthermore,compared to the NSGA-?*algorithm,the proposed TBAI algorithm achieves better Pareto fronts in four dimensions of constraints.