Task Migration and Fine Grain Parallelism on Distributed Memory Architectures
by Yvon Jegou
Abstract:
The most successful compilation techniques for distributed memory architectures are based on static analysis of the memory accesses. Loop iterations with similar comportment on the parallel memories are combined in order to form coarse grain parallel tasks. But for irregularly structured applications, the behavior of each iteration of a parallel loop on the memories is data dependent and cannot be predicted at compile-time and the only exploitable parallelism is fine-grain. We show that, because it generates parallel and asynchronous execution of a large number of small tasks, the task migration paradigm allows a direct exploitation of these irregularly structured problems on distributed memory architectures.
Keywords: software, task migration, distributed memory, fine grain, irregular code
Source:
Y. Jegou, Task Migration and Fine Grain Parallelism on Distributed Memory Architectures. In V. Malyshkin (ed.),
Parallel Computing Technologies: Proceedings of the 4th International Conference,
Lect. Notes in Comp. Sci., Vol. 1277, Springer, 1997, pp. 226-240
