Re: [OMPI users] Problem: openmpi using lam
On Mar 8, 2008, at 6:26 AM, Justus Schwabedal wrote: I thought LAM/MPI is being merged into openmpi. I think, Im not sure how the implementations are related. Think of LAM/MPI as one of the predecessors to Open MPI (several MPI projects, including LAM/MPI, merged to create the new Open MPI). So some of the ideas and code were taken from LAM, but not everything. I think lam is just a runtime for which now openmpi delivers some support. Not really. We originally created a new run-time (ORTE) for Open MPI that uses some of the ideas from LAM (and others), and a bunch of new/ original work. These days, I would only consider ORTE to be distantly related to LAM. If not so, I dont understand. However orte is another runtime right? I thought I can switch between those using something like mpi- selector-menu. mpi-selector-menu is a utility that I wrote for the OFED (OpenFabrics Enterprise Distribution) distribution. It allows you to switch between the different MPI implementations installed by OFED without requiring the user to edit their shell startup files (e.g., .bashrc). I know that the Cisco OFED distribution includes documentation on mpi- selector-menu (available from cisco.com), and there's a man page installed on mpi-selector-menu as well (and mpi-selector, the script that does the work behind mpi-selector-menu). Cisco's OFED ships with 9 different MPI implementation: Open MPI, MVAPICH, MVAPICH2. Each of those are compiled against the GNU compilers, intel compilers, and PGI compilers. mpi-selector-menu allows you to switch between these 9. I think Im lost. How can I use openmpi to run a distributed system efficiently? I dont want to fall back on old LAM/MPI since I want to be modern :o) Yours, J Original-Nachricht Datum: Fri, 7 Mar 2008 13:19:16 -0500 Von: Jeff Squyres An: Open MPI Users Betreff: Re: [OMPI users] Problem: openmpi using lam On Mar 7, 2008, at 11:31 AM, Justus Schwabedal wrote: I need advise to get openmpi to work using lam. I think you're slightly confused -- LAM/MPI and Open MPI are two different software projects that implement the same standard (MPI). This list is for supporting Open MPI. Another list (http://www.lam-mpi.org/mailman/listinfo.cgi/lam ) is available for supporting LAM/MPI. You might want to refer to the "Getting Help" page on the LAM/MPI site (http://www.lam-mpi.org/using/support/ ). I suspect that your problem has to do with your hostfile, and you should check the LAM/MPI user's guide for the answer. But be aware that LAM/MPI is in maintenance mode -- it is being deprecated in favor of Open MPI. See the announcement at the top of LAM/MPI's home page. Good luck. For testing I installed it on two machines running opensuse10.3 (both atholon 64) connected via ethernet. I then try to run a test script "hello.x" with the following commands: jus@cusp:~/Code/Spielwiese/Mpi> make hello mpicc hello.c -o hello.x jus@cusp:~/Code/Spielwiese/Mpi> lamboot mfile LAM 7.1.4/MPI 2 C++/ROMIO - Indiana University Enter passphrase for key '/cusp/jus/.ssh/id_rsa': Enter passphrase for key '/cusp/jus/.ssh/id_rsa': jus@cusp:~/Code/Spielwiese/Mpi> lamnodes n0 cusp:1:origin,this_node n1 neutrino.stat.physik.uni-potsdam.de:1: jus@cusp:~/Code/Spielwiese/Mpi> mpirun -np 2 hello.x Process 0 on cusp out of 2 Process 1 on cusp out of 2 The jobs are thus wrapped only on the node "cusp", not on both. I have no clue why this is. Thanks, Justus -- Psst! Geheimtipp: Online Games kostenlos spielen bei den GMX Free Games! http://games.entertainment.gmx.net/de/entertainment/games/free ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users -- Jeff Squyres Cisco Systems ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users -- Pt! Schon vom neuen GMX MultiMessenger gehört? Der kann`s mit allen: http://www.gmx.net/de/go/multimessenger ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users -- Jeff Squyres Cisco Systems
Re: [OMPI users] Problem: openmpi using lam
I got it. The answer I was looking for was: lam deamons are not in use anymore. mpirun does everything byitself as soon as all the path variables are set up correctly. However, what are the openIB and uDAPL warnings good for? E.g: " libibverbs: Fatal: couldn't read uverbs ABI version. -- [0,1,1]: OpenIB on host neutrino was unable to find any HCAs. Another transport will be used instead, although this may result in lower performance. -- -- [0,1,1]: uDAPL on host neutrino was unable to find any NICs. Another transport will be used instead, although this may result in lower performance. -- " Yours, Justus On Mar 8, 2008, at 4:40 PM, Reuti wrote: Am 08.03.2008 um 12:26 schrieb Justus Schwabedal: I thought LAM/MPI is being merged into openmpi. I think, Im not sure how the implementations are related. I think lam is just a runtime for which now openmpi delivers some support. If not so, I dont understand. However orte is another runtime right? I thought I can switch between those using something like mpi-selector-menu. I think Im lost. How can I use openmpi to run a distributed system efficiently? I dont want to fall back on old LAM/MPI since I want to be modern :o) Yours, J You can't mix different MPI implementations. If you compiled your program with LAM/MPI, you must use their mpirun. There is no binary interface defined for MPI, so that you could change the to be used MPI implementation just on the fly from one run to the next. If you want to change to Open MPI, you have to recompile your program and use Open MPI's mpirun. To your test-run: you specified to use two processes, but not where to start them. Hence you must provide something like a machine file. In LAM/MPI you could try: mpirun C hello.x (again: this only works in LAM/MPI). In a distributed environment it's must common to use a queuing sytem to select free nodes most efficiently for you. E.g. SGE ( http:// gridengine.sunsource.net/ ) or Torque ( http:// www.clusterresources.com/pages/products/torque-resource-manager.php ) and to start the jobs under their supervision. -- Reuti Original-Nachricht Datum: Fri, 7 Mar 2008 13:19:16 -0500 Von: Jeff Squyres An: Open MPI Users Betreff: Re: [OMPI users] Problem: openmpi using lam On Mar 7, 2008, at 11:31 AM, Justus Schwabedal wrote: I need advise to get openmpi to work using lam. I think you're slightly confused -- LAM/MPI and Open MPI are two different software projects that implement the same standard (MPI). This list is for supporting Open MPI. Another list (http://www.lam-mpi.org/mailman/listinfo.cgi/lam ) is available for supporting LAM/MPI. You might want to refer to the "Getting Help" page on the LAM/MPI site (http://www.lam-mpi.org/using/support/ ). I suspect that your problem has to do with your hostfile, and you should check the LAM/MPI user's guide for the answer. But be aware that LAM/MPI is in maintenance mode -- it is being deprecated in favor of Open MPI. See the announcement at the top of LAM/MPI's home page. Good luck. For testing I installed it on two machines running opensuse10.3 (both atholon 64) connected via ethernet. I then try to run a test script "hello.x" with the following commands: jus@cusp:~/Code/Spielwiese/Mpi> make hello mpicc hello.c -o hello.x jus@cusp:~/Code/Spielwiese/Mpi> lamboot mfile LAM 7.1.4/MPI 2 C++/ROMIO - Indiana University Enter passphrase for key '/cusp/jus/.ssh/id_rsa': Enter passphrase for key '/cusp/jus/.ssh/id_rsa': jus@cusp:~/Code/Spielwiese/Mpi> lamnodes n0 cusp:1:origin,this_node n1 neutrino.stat.physik.uni-potsdam.de:1: jus@cusp:~/Code/Spielwiese/Mpi> mpirun -np 2 hello.x Process 0 on cusp out of 2 Process 1 on cusp out of 2 The jobs are thus wrapped only on the node "cusp", not on both. I have no clue why this is. Thanks, Justus -- Psst! Geheimtipp: Online Games kostenlos spielen bei den GMX Free Games! http://games.entertainment.gmx.net/de/entertainment/games/free ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users -- Jeff Squyres Cisco Systems ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users -- Pt! Schon vom neuen GMX MultiMessenger gehört? Der kann`s mit allen: http://www.gmx.net/de/go/multimessenger ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users ___ users mailing list us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/users