A NOVEL FAULT TOLERANCE APPROACH TO SCHEDULE JOBS IN GRID ENVIRONMENT

Abstract

Grids are type of distributed machines in which aggregation of resources is done to provide services. They are present as systems which extend internet environments with machines distributed across multiple organizations and administrative domains. The power of Grid computing depends upon abundance of network connected systems and bandwidth available for computation, collaboration and communication over the network. Dependable, consistent, pervasive and inexpensive computing infrastructure is the core of Grid computing. Its virtual image provides a single point of access to powerful distributed resources. It enables the deployment of resources such as desktops, databases and storages wherever and whenever needed, as has been demonstrated by the projects such as SETI@home, MyGrid. As more and more resources are connecting over network, the grid size becomes more large, which increase the probability of failure due to network isolation, resource failure etc. To achieve high performance in grid, task must be continue its processing in the presence of faults, which means to provide a fault tolerance environment to task, so that high performance can be achieved in grid. In this dissertation, a novel fault tolerance approach is proposed. Fault tolerance approaches mainly lie into two categories, Replica Based approach and Check-Pointing approach. The main problem with Replica Based approach is its non-applicability to cost based resources whereas Check-Pointing approach suffers from inherent disadvantages of taking Check-Point, like overhead and wastage of time. Till now no one has focused on nature of tasks submitted to grid. In this dissertation, a novel approach for fault tolerance is proposed. In this approach, large task is divided into various subtasks on the basis of data flow and control flow dependencies. The experimental results show that proposed approach is efficient than Check-Point approach in terms of various parameters like number of Gridlets successfully completed and average execution time of task.