MPI collective algorithm selection and quadtree encoding
Parallel Computing
MPI Reduction Operations for Sparse Floating-point Data
Proceedings of the 15th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface
Process cooperation in multiple message broadcast
Parallel Computing
Two algorithms of irregular scatter/gather operations for heterogeneous platforms
EuroMPI'10 Proceedings of the 17th European MPI users' group meeting conference on Recent advances in the message passing interface
MPI collective algorithm selection and quadtree encoding
EuroPVM/MPI'06 Proceedings of the 13th European PVM/MPI User's Group conference on Recent advances in parallel virtual machine and message passing interface
Parallel prefix (scan) algorithms for MPI
EuroPVM/MPI'06 Proceedings of the 13th European PVM/MPI User's Group conference on Recent advances in parallel virtual machine and message passing interface
Decision trees and MPI collective algorithm selection problem
Euro-Par'07 Proceedings of the 13th international Euro-Par conference on Parallel Processing
Process cooperation in multiple message broadcast
PVM/MPI'07 Proceedings of the 14th European conference on Recent Advances in Parallel Virtual Machine and Message Passing Interface
Hi-index | 0.00 |
Collective operations are an important aspect of the currentlymost important message-passing programming modelMPI (Message Passing Interface). Many MPI applicationsmake heavy use of collective operations. Collective operationsinvolve the active participation of a known groupof processes and are usually implemented on top of MPIpoint-to-point message passing. Many optimizations of theused communication algorithms have been developed, butthe vast majority of those optimizations is still based onplain MPI point-to-point message passing. While this hasthe advantage of portability, it often does not allow for fullexploitation of the underlying interconnection network. Inthis paper, we present a low-level, pipeline-based optimizationof one-to-many and many-to-one collective operationsfor the SCI (Scalable Coherent Interface) interconnectionnetwork. The optimizations increase the performance ofsome operations by a factor of four if compared with thegeneric, tree-based algorithms.