Research On Grid Task Scheduling Using Improved Genetic Algorithm

The Grid Computing aims to provide an integrated computer platform among large number of heterogenic resources, and has becoming an important development trend for parallel and distributed computing technology. As the characters of distributed and shared, self-resembling, dynamic and diverse, autonomous and manifold management of a grid, some tasks scheduling algorithms such as Genetic Algorithm, Min-Min, Simulated Annealing etc have been presented to improve the performance of the grid, reduce the execution time and the consumption of the grid computation. Meantime, new grid simulation tools are gradually being used to help the designers of the grid systems to validate the scheme and test the performance of the grid system, and using simulation tools for the part of tasks scheduling has becoming the key point of the research in this field.In this paper, based on the introduction of the basic concepts, characters both in fields of grid and the tasks scheduling, the TITAN grid structure within a PACE performance prediction system is presented. Then, several typical grid technologies and resource management systems are discussed. Furthermore, a improved genetic algorithm (IGA) is proposed. The distinctive feature of the approach is that the IGA needs to execute only very few generations to come up with good solutions, making it suitable to be used in intense tasks environment. A fitness function with three weights used to prioritize any particular component in accordance with the needs of the jobs or the users or the resource provider is also designed to minimize makespan, idle time of the available computational resources, and the specified deadlines provided by users. It is unnecessary to use real systems to evaluate the performance and accuracy of these algorithms, so, emulators are used to handle with this kind of work, in this paper, a grid modeling and simulation toolkit named GridSim is introduced in details. Finally, GridSim is used to make the simulation of the IGA, through which, the feasibility of the algorithm is proved.It is still a first attempt to support IGA in grid tasks scheduling. The research covered several aspects without concerning the impacts from the intercommunication cost between tasks and the resources, even the actual status of the underlying networks. What's more, the scheduling frequency as well as adapt the algorithm to grid field, still needs more deliberation.