A New Genetic Algorithm For Solving Multi-objective Security Model Of Grid Independent Tasks

Grid task scheduling is an important part of grid computing. It directly affects theperformance of grid computing system. However, due to the characteristics of the gridenvironment, such as, heterogeneity, distributed, openness, uncertainty and dynamics,it presents new challenges to the traditional scheduling strategy. For the schedulingproblems under the grid environment, we should not only consider the time cost ofperformance, but also should consider new issues arising from the characteristics of thegrid, such as the credibility of the grid. In this thesis, a new model for independenttasks security scheduling is constructed and a genetic algorithm is used to solve it. Themain contributions of this thesis are summarized as follows:1. For the independent tasks scheduling problems, a number of factors, such as thecritical time of scheduling, the performance of security, the credibility, and the cost ofscheduling are taken into account as objectives. The critical time of tasks scheduling isconsidered as the first objective, the performance of security in the tasks scheduling isconsidered as the second objective, the credibility of the grid nodes is considered as thethird objective, and the cost of tasks scheduling is considered as the fourth objective.The constraint is that the needed security of every task should not exceed the securityof the computer assigned.This thesis mainly focuses on improving the critical time of tasks scheduling andthe performance of security. The critical time of tasks scheduling, which is a mini-maxproblem, is transformed into a two objective optimization problem, in which the tasksscheduling make span and load balance are optimized. Then the four objectiveoptimization model is transformed into a five objective optimization model, whichhighlights the greater practical significance. Finally, the constrained multi-objectiveoptimization problem is converted into a unconstrained multi-objective optimizationproblem by the penalty function method.2. We improve the performance of NSGA-II, making it suitable for the designedmodel. In order to enhance the diversity of the population, we reduce the selectionprobability of bad solutions and add some randomly selected individuals. Based on allof these, the elite selection strategy in the NSGA-II is improved. Secondly, we design aspecial crossover operator and mutation operator in order to enhance the search width.Thirdly, the simulation is carried out and the convergence of the proposed algorithm is proved as well. Finally, the results show that the improved NSGA-II is moreeffective than the compared ones.