Achieving Guaranteed Service with Fault-Tolerant Resources in Grid

Composed of loosely coupled virtual resources, grid, being highly distinguished from traditional high-performance computing, is extensively used in computation-intensive problem solving in the arenas of science and technology. Maintaining performance or balancing load of each resource in grid is always more challenging with high chances of resource failure. The objective of this paper is to improve the efficiency of the Nearest Deadline First Scheduled (NDFS) algorithm considering resource failure a sudden occurrence in grid. The algorithm introduces periodical runtime backup to another available resource for retaining Quality-of-Service as approved in service quality agreement. This paper presents multiple job execution cases through implementation of benchmark codes executed in local grid test bed using Globus Toolkit middleware, with an emphasis on resource failure phenomenon of grid. These experimental results establish the requirements of the proposed algorithm to ensure the job deadline misses get reduced even if unexpected resource failures happen.