Newsgroups: comp.parallel.mpi
From: mst@ZPR.Uni-Koeln.DE (Martin Strietzel)
Subject: Datatranspositon by MPI_ALLTOALL
Organization: Institute for Mathematics, University of Cologne, Germany
Date: 4 Jul 1996 09:22:19 GMT
Message-ID: <4rg2gb$pbc@news.rrz.uni-koeln.de>

Hi,

I am using the MPI_ALLTOALL procedure for doing a data-transpositon 
on a two- or one-dimensional gridtopology. 
Can anybody give me some information about which algorithm the 
programmers of the different MPI implementations (MPICH, Cray MPI, 
IBM MPI, SGI MPI) use for this all-to-all call? Do they optimize it 
for the different hardware topologies? Has anybody benchmarked a 
datatransposition on different machines with different MPI 
implementations?

Any comments and citations are welcome!

Thanks,
	Martin Strietzel

-- 
-------------------------------------------------------------------

  Martin Strietzel                   

  Zentrum fuer Paralleles Rechnen / DLR     
  Weyertal 80            
  50931 Koeln - Lindenthal     e-Mail: strietzel@zpr.uni-koeln.de

  Tel.: +49 (0)221/470-6003      Fax : +49 (0)221/470-5160
     or +49 (0)221/470-6030            	

           WWW: http://www.zpr.uni-koeln.de/~mst/
-------------------------------------------------------------------

