The performance of two dynamic, distributed load balancing methods for small-grained tasks on large parallel machines are compared. The contracting within neighborhood (CWN) approach schedules a task on some processing element (PE) as soon as it is created; the gradient model (GM) keeps newly created tasks on the source PE, and distributes them when required. It is found that although CWN performs better than GM in most of the tests performed, both techniques fall significantly short of theoretical optimum performance.
|Original language||English (US)|
|Number of pages||5|
|Journal||Proceedings of the International Conference on Parallel Processing|
|State||Published - Dec 1 1988|
ASJC Scopus subject areas
- Hardware and Architecture