Hi Stefano That doesn't seem to be an Open MPI problem, but an Intel environment problem.
Here I have this library directory (on a slightly older version): .../composerxe/2013.3.163/compiler/lib/intel64/ but Intel keeps changing their directory structure, playing with bunches of soft links, etc, (which is quite annoying), so your newer version may not have the same directories. However, this should be all properly set if you run the Intel "source compilervars.[sh,csh] intel64", as Jeff noted. I read on your first email that you do that. You don't need/shouldn't to change that by hand, unless Intel really messed up their own environment settings on their compilervars script. Did you do "ldd your_executable", to see if it finds libimf.so ? I hope this helps, Gus Correa On 06/21/2013 02:05 PM, Stefano Zaghi wrote:
Hi Gus, thank you for your replay. The strange path I have chosen is because this was only a test. However my home dir is shared on all nodes and the lib dir is not a simple simlink. I think that Thomas is right, I have to remove intel64 from Intel/lib path. Monday I will try. Thank you again. Il giorno 21/giu/2013 17:55, "Gus Correa" <[email protected] <mailto:[email protected]>> ha scritto: Hi Stefano Make sure your Intel compiler's shared libraries are accessible on all nodes. Is your /home directory shared across all nodes? How about /opt (if Intel is installed there)? By default Intel installs the compilers on /opt, which in typical clusters (and Linux distributions) is a local directory (to each node), not shared via NFS. Although you seem to have installed it somewhere else, /home/stefano/opt maybe, if /home/stefano/opt is just a soft link to /opt, not a real directory, that may not to do the trick across the cluster network. This error: >> /home/stefano/opt/mpi/openmpi/__1.6.4/intel/bin/orted: error >> while loading shared libraries: libimf.so: cannot open shared >> object file: No such file or directory suggests something like that is going on (libimf.so is an *Intel shared library*, it is *not an Open MPI libary*). To have all needed tools (OpenMPI and Intel) available on all nodes, there are two typical solutions (by the way, see this FAQ: http://www.open-mpi.org/faq/?__category=building#where-to-__install <http://www.open-mpi.org/faq/?category=building#where-to-install>): 1) Install them on all nodes, via RPM, or configure/make/install, or other mechanism. This is time consuming and costly to maintain, but scales well in big or small clusters. 2) Install them on your master/head/adminsitration/__storage node, and and share them via network (typicaly via NFS export/mount). This is easy to maintain, and scales well in small/medium clusters, but not so much on big ones. Make sure the Intel and MPI directories are either shared by or present/installed on all nodes. I also wonder if you really need these many environment variables: >> LD_LIBRARY_PATH=${MPI}/lib/__openmpi:${MPI}/lib:$LD___LIBRARY_PATH >> export LD_RUN_PATH=${MPI}/lib/__openmpi:${MPI}/lib:$LD_RUN___PATH or if that may be actually replaced by the simpler form: >> LD_LIBRARY_PATH=${MPI}/lib:$__LD_LIBRARY_PATH I hope it helps, Gus Correa On 06/21/2013 04:35 AM, Stefano Zaghi wrote: Wow... I think you are right... I will am check after the job I have just started will finish. Thank you again. See you soon Stefano Zaghi Ph.D. Aerospace Engineer, Research Scientist, Dept. of Computational Hydrodynamics at *CNR-INSEAN* <http://www.insean.cnr.it/en/__content/cnr-insean <http://www.insean.cnr.it/en/content/cnr-insean>> The Italian Ship Model Basin (+39) 06.50299297 <tel:%28%2B39%29%2006.50299297> (Office) My codes: *OFF* <https://github.com/szaghi/OFF__>, Open source Finite volumes Fluid dynamics code *Lib_VTK_IO* <https://github.com/szaghi/__Lib_VTK_IO <https://github.com/szaghi/Lib_VTK_IO>>, a Fortran library to write and read data conforming the VTK standard *IR_Precision* <https://github.com/szaghi/IR___Precision <https://github.com/szaghi/IR_Precision>>, a Fortran (standard 2003) module to develop portable codes 2013/6/21 <[email protected] <mailto:[email protected]> <mailto:thomas.forde@ulstein.__com <mailto:[email protected]>>> hi Stefano /home/stefano/opt/intel/2013.__4.183/lib/intel64/ is also the wrong path, as the file is in ..183/lib/ and not ...183/lib/intel64/ is that why? ./Thomas Den 21. juni 2013 kl. 10:26 skrev "Stefano Zaghi" <[email protected] <mailto:[email protected]> <mailto:stefano.zaghi@gmail.__com <mailto:[email protected]>>>: Dear Thomas, thank you again. Symlink in /usr/lib64 is not enough, I have symlinked also in /home/stefano/opt/mpi/openmpi/__1.6.4/intel/lib and, as expected, not only libimf.so but also ibirng.so and libintlc.so.5 are necessary. Now also remote runs works, but this is only a workaround, I still not understand why mpirun do not find intel library even if LD_LIBRARY_PATH contains also /home/stefano/opt/intel/2013.__4.183/lib/intel64. Can you try explain again? Thank you very much. Stefano Zaghi Ph.D. Aerospace Engineer, Research Scientist, Dept. of Computational Hydrodynamics at *CNR-INSEAN* <http://www.insean.cnr.it/en/__content/cnr-insean <http://www.insean.cnr.it/en/content/cnr-insean>> The Italian Ship Model Basin (+39) 06.50299297 <tel:%28%2B39%29%2006.50299297> (Office) My codes: *OFF* <https://github.com/szaghi/OFF__>, Open source Finite volumes Fluid dynamics code *Lib_VTK_IO* <https://github.com/szaghi/__Lib_VTK_IO <https://github.com/szaghi/Lib_VTK_IO>>, a would Fortran library to write and read data conforming the VTK standard *IR_Precision* <https://github.com/szaghi/IR___Precision <https://github.com/szaghi/IR_Precision>>, a Fortran (standard 2003) module to develop portable codes 2013/6/21 <[email protected] <mailto:[email protected]> <mailto:thomas.forde@ulstein.__com <mailto:[email protected]>>> your settings are as following: export MPI=/home/stefano/opt/mpi/__openmpi/1.6.4/intel export PATH=${MPI}/bin:$PATH export LD_LIBRARY_PATH=${MPI}/lib/__openmpi:${MPI}/lib:$LD___LIBRARY_PATH export LD_RUN_PATH=${MPI}/lib/__openmpi:${MPI}/lib:$LD_RUN___PATH and your path to libimf.so file is /home/stefano/opt/intel/2013.__4.183/lib/libimf.so your export LD_LIbrary_PATH if i can decude it right would be because you use the $MPI first. /home/stefano/opt/mpi/openmpi/__1.64./intel/lib/openmpi and /home/stefano/opt/mpi/openmpi/__1.64./intel/lib as you can see it doesnt look for the files int he right place. the simplest thing i would try is to symlink the libimf.so file to /usr/lib64 and should give you a workaround. From: Stefano Zaghi <[email protected] <mailto:[email protected]> <mailto:stefano.zaghi@gmail.__com <mailto:[email protected]>>> To: Open MPI Users <[email protected] <mailto:[email protected]> <mailto:[email protected] <mailto:[email protected]>>>, Date: 21.06.2013 09 <tel:21.06.2013%2009>:45 Subject: Re: [OMPI users] OpenMPI 1.6.4 and Intel Composer_xe_2013.4.183: problem with remote runs, orted: error while loading shared libraries: libimf.so Sent by: [email protected] <mailto:[email protected]> <mailto:users-bounces@open-__mpi.org <mailto:[email protected]>> ------------------------------__------------------------------__------------ Dear Thomas, thank you very much for your very fast replay. Yes I have that library in the correct place: -rwxr-xr-x 1 stefano users 3.0M May 20 14:22 opt/intel/2013.4.183/lib/__intel64/libimf.so Stefano Zaghi Ph.D. Aerospace Engineer, Research Scientist, Dept. of Computational Hydrodynamics at *_CNR-INSEAN_* <http://www.insean.cnr.it/en/__content/cnr-insean <http://www.insean.cnr.it/en/content/cnr-insean>> The Italian Ship Model Basin (+39) 06.50299297 <tel:%28%2B39%29%2006.50299297> (Office) My codes: *_OFF_* <https://github.com/szaghi/OFF__>, Open source Finite volumes Fluid dynamics code *_Lib_VTK_IO_* <https://github.com/szaghi/__Lib_VTK_IO <https://github.com/szaghi/Lib_VTK_IO>>, a Fortran library to write and read data conforming the VTK standard *_IR_Precision_* <https://github.com/szaghi/IR___Precision <https://github.com/szaghi/IR_Precision>>, a Fortran (standard 2003) module to develop portable codes 2013/6/21 <[email protected]_ <mailto:thomas.forde@ulstein.__com <mailto:[email protected]>>> hi Stefano your error message show that you are missing a shared library, not necessary that library path is wrong. do you actually have libimf.so, can you find the file on your system. ./Thomas From: Stefano Zaghi <[email protected]_ <mailto:stefano.zaghi@gmail.__com <mailto:[email protected]>>> To: [email protected]_ <mailto:[email protected] <mailto:[email protected]>>, Date: 21.06.2013 09 <tel:21.06.2013%2009>:27 Subject: [OMPI users] OpenMPI 1.6.4 and Intel Composer_xe_2013.4.183: problem with remote runs, orted: error while loading shared libraries: libimf.so Sent by: [email protected]_ <mailto:users-bounces@open-__mpi.org <mailto:[email protected]>> ------------------------------__------------------------------__------------ Dear All, I have compiled OpenMPI 1.6.4 with Intel Composer_xe_2013.4.183. My configure is: ./configure --prefix=/home/stefano/opt/__mpi/openmpi/1.6.4/intel CC=icc CXX=icpc F77=ifort FC=ifort Intel Composer has been installed in: /home/stefano/opt/intel/2013.__4.183/composer_xe_2013.4.183 Into the .bashrc and .profile in all nodes there is: source /home/stefano/opt/intel/2013.__4.183/bin/compilervars.sh intel64 export MPI=/home/stefano/opt/mpi/__openmpi/1.6.4/intel export PATH=${MPI}/bin:$PATH export LD_LIBRARY_PATH=${MPI}/lib/__openmpi:${MPI}/lib:$LD___LIBRARY_PATH export LD_RUN_PATH=${MPI}/lib/__openmpi:${MPI}/lib:$LD_RUN___PATH If I run parallel job into each single node (e.g. mpirun -np 8 myprog) all works well. However, when I tried to run parallel job in more nodes of the cluster (remote runs) like the following: mpirun -np 16 --bynode --machinefile nodi.txt -x LD_LIBRARY_PATH -x LD_RUN_PATH myprog I got the following error: /home/stefano/opt/mpi/openmpi/__1.6.4/intel/bin/orted: error while loading shared libraries: libimf.so: cannot open shared object file: No such file or directory I have read many FAQs and online resources, all indicating LD_LIBRARY_PATH as the possible problem (wrong setting). However I am not able to figure out what is going wrong, the LD_LIBRARY_PATH seems to set right in all nodes. It is worth noting that in the same cluster I have successful installed OpenMPI 1.4.3 with Intel Composer_xe_2011_sp1.6.233 following exactly the same procedure. Thank you in advance for all suggestion, sincerely Stefano Zaghi Ph.D. Aerospace Engineer, Research Scientist, Dept. of Computational Hydrodynamics at *_CNR-INSEAN_* <http://www.insean.cnr.it/en/__content/cnr-insean <http://www.insean.cnr.it/en/content/cnr-insean>> The Italian Ship Model Basin (+39) 06.50299297 <tel:%28%2B39%29%2006.50299297> (Office) My codes: _ _*_OFF_* <https://github.com/szaghi/OFF__>, Open source Finite volumes Fluid dynamics code _ _*_Lib_VTK_IO_* <https://github.com/szaghi/__Lib_VTK_IO <https://github.com/szaghi/Lib_VTK_IO>>, a Fortran library to write and read data conforming the VTK standard *_IR_Precision_* <https://github.com/szaghi/IR___Precision <https://github.com/szaghi/IR_Precision>>, a Fortran (standard 2003) module to develop portable codes_________________________________________________ users mailing list_ [email protected]_ <mailto:[email protected] <mailto:[email protected]>>_ __http://www.open-mpi.org/__mailman/listinfo.cgi/users_ <http://www.open-mpi.org/mailman/listinfo.cgi/users_> Denne e-posten kan innehalde informasjon som er konfidensiell og/eller underlagt lovbestemt teieplikt. Kun den tiltenkte adressat har adgang til å lese eller vidareformidle denne e-posten eller tilhøyrande vedlegg. Dersom De ikkje er den tiltenkte mottakar, vennligst kontakt avsendar pr e-post, slett denne e-posten med vedlegg og makuler samtlige utskrifter og kopiar av den. This e-mail may contain confidential information, or otherwise be protected against unauthorised use. Any disclosure, distribution or other use of the information by anyone but the intended recipient is strictly prohibited. If you have received this e-mail in error, please advise the sender by immediate reply and destroy the received documents and any copies hereof. PBefore printing, think about the environment _________________________________________________ users mailing list_ [email protected]_ <mailto:[email protected] <mailto:[email protected]>>_ __http://www.open-mpi.org/__mailman/listinfo.cgi/users_ <http://www.open-mpi.org/mailman/listinfo.cgi/users_> _________________________________________________ users mailing list [email protected] <mailto:[email protected]> <mailto:[email protected] <mailto:[email protected]>> http://www.open-mpi.org/__mailman/listinfo.cgi/users <http://www.open-mpi.org/mailman/listinfo.cgi/users> Denne e-posten kan innehalde informasjon som er konfidensiell og/eller underlagt lovbestemt teieplikt. Kun den tiltenkte adressat har adgang til å lese eller vidareformidle denne e-posten eller tilhøyrande vedlegg. Dersom De ikkje er den tiltenkte mottakar, vennligst kontakt avsendar pr e-post, slett denne e-posten med vedlegg og makuler samtlige utskrifter og kopiar av den. This e-mail may contain confidential information, or otherwise be protected against unauthorised use. Any disclosure, distribution or other use of the information by anyone but the intended recipient is strictly prohibited. If you have received this e-mail in error, please advise the sender by immediate reply and destroy the received documents and any copies hereof. PBefore printing, think about the environment _________________________________________________ users mailing list [email protected] <mailto:[email protected]> <mailto:[email protected] <mailto:[email protected]>> http://www.open-mpi.org/__mailman/listinfo.cgi/users <http://www.open-mpi.org/mailman/listinfo.cgi/users> _________________________________________________ users mailing list [email protected] <mailto:[email protected]> <mailto:[email protected] <mailto:[email protected]>>htt__p://www.open-mpi.org/mailman/__listinfo.cgi/users <http://www.open-mpi.org/mailman/listinfo.cgi/users> Denne e-posten kan innehalde informasjon som er konfidensiell og/eller underlagt lovbestemt teieplikt. Kun den tiltenkte adressat har adgang til å lese eller vidareformidle denne e-posten eller tilhøyrande vedlegg. Dersom De ikkje er den tiltenkte mottakar, vennligst kontakt avsendar pr e-post, slett denne e-posten med vedlegg og makuler samtlige utskrifter og kopiar av den. This e-mail may contain confidential information, or otherwise be protected against unauthorised use. Any disclosure, distribution or other use of the information by anyone but the intended recipient is strictly prohibited. If you have received this e-mail in error, please advise the sender by immediate reply and destroy the received documents and any copies hereof. PBefore printing, think about the environment _________________________________________________ users mailing list [email protected] <mailto:[email protected]> <mailto:[email protected] <mailto:[email protected]>> http://www.open-mpi.org/__mailman/listinfo.cgi/users <http://www.open-mpi.org/mailman/listinfo.cgi/users> _________________________________________________ users mailing list [email protected] <mailto:[email protected]> http://www.open-mpi.org/__mailman/listinfo.cgi/users <http://www.open-mpi.org/mailman/listinfo.cgi/users> _________________________________________________ users mailing list [email protected] <mailto:[email protected]> http://www.open-mpi.org/__mailman/listinfo.cgi/users <http://www.open-mpi.org/mailman/listinfo.cgi/users> _______________________________________________ users mailing list [email protected] http://www.open-mpi.org/mailman/listinfo.cgi/users
