Research On Scheduling Of Grid Load Balancing Based On Swarm Intelligence

Grid Computing [1] is an emerging computing model which provides the ability to perform higher throughput computing by taking advantage of many networked computers and distributing process execution across a parallel infrastructure. The huge amount of computations a grid can fulfill in a specific time, security, reliability, and accuracy cannot be done by the best super computers. When grid resources are required by lots of tasks, the system can optimize the resources only by scheduling the tasks reasonably. A good task scheduling can adapt the dynamic environment of the gird flexibility, reduce the time required to complete all tasks, use the resources in the grid efficiently. Grid performance can still be improved by making sure that all the resources available in the Grid are utilized by a proper load balancing algorithm.By analyzing of characteristics of the task scheduling algorithm in grid environment, the advantages and disadvantages of the traditional task scheduling algorithms are thoroughly studied and compare. This paper proposes two new algorithms based on swarm intelligence, the Particle Swarm Optimization based on centralized approach (CPSO) and Particle Swarm Optimization based on distributed approach (DPSO). Both algorithms are based on the traditional particle swarm algorithm, with the position and speed of particles redefined. The CPSO algorithm combines the advantages of traditional centralized algorithm, solving the problem of the continuity of the algorithm and solution of discrete conflict. The DPSO algorithm redefines the basic concept of the PSO algorithm, combines with behavior of the intelligent swarm features and characteristics of the resources to handle tasks in grid computing, adapts to the grid environment with the characteristic of dynamic and scalability.The proposed approaches are simulated by using a Grid simulation toolkit (GridSim) dedicated to Grid simulations. The performance of the algorithms is evaluated using several performance criteria. Experiments are conducted by setting different parameters, changing the characteristics of tasks and resources in the grid to study the algorithm on the scalability, simplicity and adaptability. The simulation results of the proposed approaches are compared with a classical approach called State Broadcast Algorithm and random approach, showing that the proposed algorithms can perform very well in a Grid environment.