Title :
An evaluation of two implementation strategies for optimizing one-sided atomic reduction
Author :
Nieplocha, Jarek ; Tipparaju, Vinod ; Apra, Edoardo
Author_Institution :
Pacific Northwest Nat. Lab., Richland, WA, USA
Abstract :
One-sided atomic reduction, also known as the accumulate operation, combines atomically a content of the local buffer with data at remote memory location. This operation has been included in the MPI-2 standard a, MPI_Accumulate. The current paper discusses two strategies for implementing one-sided atomic reduction called owner-computes and caller-computes. Performance of these two schemes has been investigated on the HP AlphaServer SC45 and HP zx-2600 clusters both equipped with the Quadrics Elan-3 network.
Keywords :
message passing; performance evaluation; storage management; HP AlphaServer SC45; HP zx-2600; MPI-2 standard; Quadrics Elan-3 network; local buffer; one-sided atomic reduction; performance evaluation; remote memory location; Bandwidth; Concurrent computing; Distributed computing; Distributed processing; Laboratories; Linux; Network interfaces; Protection; Size measurement; Switches;
Conference_Titel :
Parallel and Distributed Processing Symposium, 2005. Proceedings. 19th IEEE International
Print_ISBN :
0-7695-2312-9
DOI :
10.1109/IPDPS.2005.96
