Workflow Scheduling With Deadline Constraints In Multi-Modal Cloud Service

Cloud computing is market-oriented, and provides high-quality, efficient information ser-vice for users, considering the problem of workflow scheduling with deadline constraints in multi-modal cloud service, and improving resource utilization while reducing cost, has impor-tant theoretical and practical value.Characteristics are analyzed and a mathematics model of workflow scheduling with dead-line constraints in multi-modal cloud service is established. A meta-heuristic named Iterated It-erated Integrated Local Search(IILS) is proposed for the problem, it mainly consists of four part-s, namely, the initial solution, iterative local search, perturbation, new starting point selection. Four kinds of initial solution generation methods are constructed, namely, optimal modal chosen algorithm(OPT), decrease resource algorithm(DR), decrease resource fair algorithm(DRF), in-crease resource fair(IRF). INS(insertion neighborhood structure) and SNS(swapping neighbor-hood structure) are two neighborhoods of iterated local search, and the neighborhood structure will be changed to another during the search procedure according to the corresponding con-dition. In order to prevent the algorithm into local optimum, perturbation processes which is based on a certain probability of insertion or swap operation will be introduced to increase the diversification. Considering the balance of distance factors and target factors, the new starting point selection algorithm will choose one as the initial solution of next iteration according to certain acceptance criteria from a solution set which is produced by the perturbation processes.Through experimental testing and the multivariate analysis method (ANOVA), the initial solution generation, local search, perturbation, new starting point selection and other relevant parameters are tested and analyzed, and the optimal parameter combination is obtained. By comparison with the existing algorithms on a large number of instances, the effectiveness of the proposed algorithm was proved.