In this paper, we consider the parallel implementation of a block Cholesky factorization based on a nested dissection ordering for unstructured problems. We focus on loosely coupled networks of many processors with local memory and message passing mechanism. More precisely, we study a parallel block solver associated with refined partitions from the separator partition; the aim is to find the partition corresponding to the correct granularity leading to a high quality mapping (in terms of load balancing for the processors, of average length for the routing paths, and of average edge contention on the network). Then, we propose a refinement algorithm leading to this good granularity, and we provide some numerical measurements using the mapping tool included in the ADAM environment.
CITATION STYLE
Charrier, P., & Roman, J. (1992). Partitioning and mapping for parallel nested dissection on distributed memory architectures. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 634 LNCS, pp. 295–306). Springer Verlag. https://doi.org/10.1007/3-540-55895-0_424
Mendeley helps you to discover research relevant for your work.