[Beowulf] MPICH vs MPICH2
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Rob Ross rross at mcs.anl.govWed Apr 13 10:53:00 PDT 2005
- Previous message: [Beowulf] MPICH vs MPICH2
- Next message: [Beowulf] Call for participation: CI6016, HPC-Asia, GCC, and NPC 2005
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
MPICH2 does not support heterogeneous collections of architectures as MPICH does; that's the only reason why one wouldn't want to switch. Regards, Rob Robert Latham wrote: > On Thu, Apr 07, 2005 at 03:30:09PM +0800, John Lau wrote: > >>Hi, >> >>Does anyone compare the different between MPICH and MPICH2? What are the >>greatest new features in MPICH2? And does the MPICH2 have large >>performance increase than MPICH? I would like to see some benchmarks >>that compare them. > > > Well, the foremost differnce between MPICH and MPICH2 is that MPICH2 > implementes MPI-2, whereas MPICH implmented only MPI-1 and MPI-IO. > > MPICH2 has improved datatype handling routines > > R. Ross, N. Miller, and W. Gropp, "Implementing Fast and > Reusable Datatype Processing," Proceedings of EuroPVM/MPI > 2003, Venice, Italy, September, 2003. > > MPICH2 has excellent performance for one-sided opeerations. Hard to > compare with MPICH on this one, but there are benchmarks where MPICH2 > compares favorably with other MPI-2 implementations. > > R. Thakur, W. Gropp, and B. Toonen, "Minimizing > Synchronization Overhead in the Implementation of MPI > One-Sided Communication", Proceedings of the 11th European > PVM/MPI Users' Group Conference (Euro PVM/MPI 2004), September > 2004. > > I can't find the data to back this up but MPICH2 does perform better > than MPICH1 in point-to-point messaging (as measured by NetPIPE). 28% > faster for an 8MB message. > > I hope that's enough to entice you into trying MPICH2. I work with > the MPICH2 folks, so I'm a little biased, but it's a quality > implementation. > > ==rob >
- Previous message: [Beowulf] MPICH vs MPICH2
- Next message: [Beowulf] Call for participation: CI6016, HPC-Asia, GCC, and NPC 2005
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
