Research On The Strategy Of Grid Resource Scheduling Based On Ant Colony Optimization

Grid,which is a newly-up information technology,is a great revolution on science and technology after the Internet and Web.Computing,storage space and other resources with high performance can be acquired more conveniently and economically than ever before with the help of grid.However,traditional methods for resource management are unsuitable for grid environment,because grid resources are of wide variety,heterogeneous,geographically distributed in wide areas, managed and owned by different individuals or organizations with different policies, and always changing dynamically.Now,it's urgent to study and design a new resource management system with scheduling policy for grid environment,which is an important research direction.On the other hand,since bionics was found in 1950s,with the inspiration of organic evolution,people have simulated mechanism of nature and advanced many new methods for resolving combinatorial optimization problems,such as genetic algorithm,simulated annealing,ACO(Ant Colony Optimization),PSO(Particle Swarm Optimization)and so on.Especially,ACO and PSO are both promising population-based intelligent algorithms.The former simulates the way in which natural ants search for food,and the algorithm has the characteristics of positive feedback and stability.The latter simulates the way in which natural bird colonies search for food.Later,it gradually turns into an optimization technique with high performance.In this thesis,the basic theory of grid and its relevant resource management system is introduced.The basic principle of ACO and PSO is analyzed in details.A resource allocating and task scheduling strategy,which is based on a hybrid behavior of ACO,is designed in the grid model of resource broker,in which QoS restraints(user time deadline and budget)are taken into consideration.The designed strategy consists of two steps in the preprocessing stage:at former step the PSO quickly produces a large number of solutions with high quality and transforms them into initial pheromone to start ACO.At latter step,ACO converges on global-best solution with its efficient feedback,in which a local search is also introduced to speed up the process.Finally,the strategy allocates tasks to the resources according to the best solution that is produced during the preprocessing stage.The strategy is implemented in Java and tested on the platform,which is built up with GridSim to simulate the scheduling process of grid resources.The result shows that the strategy is feasible:Compared with ACO and PSO,the strategy notably reduces the total time that tasks execute on resources,and improves the scheduling performance.