Output details
11 - Computer Science and Informatics
University of St Andrews
Granularity-Aware Work-Stealing for Computationally-Uniform Grids
<10>CCGrid is a leading conference in Grid and Cloud computing.
This paper describes novel work-stealing approaches to scheduling tasks in a wide-area (grid, cloud) setting that improve scheduling performance over the best previous techiques, such as First-Come First-Served scheduling. The paper undertakes both simulation and real experiments to verify the effectiveness of the approach, showing that major improvements (up to 20% in total execution time) can be achieved using the new granularity-aware approach. Using a new statistical measure of irregularity, it shows that these improvements correlate with the irregularity of the application.