[Beowulf] MPICH vs. OpenMPI
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.
Philippe Blaise philippe.blaise at cea.frThu Apr 24 15:39:33 PDT 2008
- Previous message: [Beowulf] MPICH vs. OpenMPI
- Next message: [Beowulf] MPICH vs. OpenMPI
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Prentice Bisbal a écrit : > Can anyone give me a quick comparison of OpenMPI vs. MPICH? I've always > used MPICH (I never liked lam), and just recently heard about OpenMPI. > Anyone here using it? > > I use OpenMPI since a few months and run programs like Abinit, Siesta with success every days (on a small IB cluster - 100 opteron cores under linux). I have done some comparisons using M(VA)PICH(2) on IB instead of OpenMPI and haven't seen any great differences ; the performances are comparable, for example on a Alltoall, depending on the message size, sometimes MVAPICH is better, sometimes OpenMPI is better. Also, I have to said that OpenMPI is quite appealing because of its ease of use, a very good web site (FAQs, mailing list), and has several advantages : it is well integrated with a PBS-like batch system, a same binary can run on different networks (IB, ethernet) without recompiling or reliking your application, and there are some possibilities to exploit numa placement of the tasks, on these last two points I'm not sure that MPICH can do the same. But, as the other guys have said before, it is very useful to have the two implementations on the same cluster. Another thing : my own experience shows that ISVs usually send you a binary which is compatible with mpich or mpich2, but not with OpenMPI. Bye Phil.
- Previous message: [Beowulf] MPICH vs. OpenMPI
- Next message: [Beowulf] MPICH vs. OpenMPI
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
