M. Frumkin and R. Hood (USA)
computational grids, benchmarks, perfor mance, navigation, dynamic scheduling.
Navigation or dynamic scheduling of applications on computational grids can be improved through the use of dy namic characterization of grid resources. We use the NAS Grid Benchmarks to map performance of grid resources into a "GridScape" representing the dynamic state of the grid. Then we use the GridScape for automatic assign ment of the tasks to grid resources. For a moderate num ber of machines in our experiments, the scalability of the scheduling system was achieved by limiting the mapping and the scheduling overhead to a few percent of the appli cation resource requirements and by distributing the Grid Scape across the grid. The use of the GridScape in task submission and assignment process guarantees that the ap plication tasks do not congest the grid. We show that our approach reduces the turn-around time of a data mining ap plication and of a flow simulation application by 25-35%.
Important Links:
Go Back
