Towards a scalable broadcast in wormhole-switched mesh networks
Proceedings of the 2002 ACM symposium on Applied computing
Towards scalable collective communication for multicomputer interconnection networks
Information Sciences: an International Journal - Special issue: Information technology
A plane-based broadcast algorithm for multicomputer networks
Journal of Systems Architecture: the EUROMICRO Journal
Performance of deterministic and adaptive broadcast algorithms in multicomputer networks
International Journal of High Performance Computing and Networking
Hi-index | 0.00 |
This work presents a methodology for cosntructing efficient collective communication libraries for mesh-based multicomputers with wormhole routing. Such libraries can be built from a small set of simple primitives. Specifically, two classes of routines are presented: short vector routines that minimize message startups and long vector routines that minimize data transmission. Routines from these two classes can then be hybridized, resulting in algorithms which are efficient over a wide range of machine performance and vector size parameters. Implementations of these basic routines are given, and the derivation of other routines from them is demonstrated. Furthermore, these methods are applicable to meshes in their full generality: There are no restrictions on the relative dimensions of the mesh or the number of nodes contained therein (e.g., they do not require power-of-two partition sizes). Also, the schemes presented here inherently support group communications as well as full-partition operations. Performance data from the Intel Paragon are included.