[molpro-user] Installation issues when installing for a cluster with a Myrinet network
Kirk Peterson
kipeters at wsu.edu
Fri May 9 17:48:48 BST 2008
Dear Evgeniy,
are you able to login to a compute node without a password? Have you
run other parallel jobs besides Molpro, i.e., can you successfully run
the test program from global arrays?
-Kirk
On May 9, 2008, at 9:40 AM, Evgeniy Gromov wrote:
> Dear Kirk and Gerald Knizia
>
> Thanks a lot for your responses. Indeed some lapack and blas libraries
> were missing on the cluster nodes and I had to set up additional
> ld_library paths. But unfortunately this had not solved the problem -
> it doesn't run anyway. Maybe I do (or did) something wrong, but when
> I run it with -v option (i.e. molpro -v) it gives:
>
> cd /tmp/egromov/
> long output file: /home/thor2/egromov/prog/molpro2006.1/
> pCMe_H2O_vert_6-31G.log
> /opt/mpichgm-intel_NEW/bin/mpirun.ch_gm -np 1 --gm-kill 10 /home/
> thor2/egromov/prog/molpro2006.1/bin/
> molprop_2006_1_i4_athlon_myrinet.exe
> odin.cl1: Connection refused
> status=256
>
> and then stops.
>
> Does it look reasonable until "odin.cl1: Connection refused"? What
> this "odin.cl1: Connection refused" with "status=256" means I have no
> idea. Likely that is the problem. Note that I get this when running
> just on one node interactively (the node has 2 processors).
>
> Best,
> Evgeniy
>
> Gerald Knizia wrote:
>> You may need to include additional dynamic libraries either in the
>> rpath when linking the executable or to the environment variable
>> LD_LIBRARY_PATH. Both are ways to tell the dynamic linker where to
>> find .so libraries. You can find out which exactly need to be
>> found by using ldd molpro.exe
>> In the output of this command, each .so required should point to a
>> concrete file on the file system with that name. You can include
>> this command in your submit-script and inspect the output in order
>> to see which are the missing libraries when molpro is executed by
>> the queing engine.
>> In the case where some external launcher (like "parallel") is
>> required, you may also need to make sure that this one can be
>> executed (try to run it and see if the OS complains).
>> Examples for a library specification with rpath (in this case for
>> the ACML math library, you will most likely need other libraries!
>> Which that are is what ldd can tell you.):
>> BLASLIB=-L/software/ACML-gnu64/lib/ -lacml -lacml_mv -lg2c -Wl,-
>> rpath,/software/ACML-gnu64/lib/
>> Here the -L tells the static linker where to find the .lib files,
>> and the Wl,-rpath tells it to embed the actual directories into the
>> executable, so that the dynamic linker (which is run when the
>> program is) can find them.
>> Such linker flags are specified at link time (i.e., when you build
>> molpro)
>> Example for LD_LIBRARY_PATH:
>> bash: export LD_LIBRARY_PATH=/software/ACML-gnu64/lib/:
>> $LD_LIBRARY_PATH
>> csh: setenv LD_LIBRARY_PATH /software/ACML-gnu64/lib/:
>> $LD_LIBRARY_PATH
>> these need to be specified before molpro is run. (i.e., in some
>> shell-script that is run before molpro is invoked).
>
>
> Kirk Peterson wrote:
>> Dear Evgeniy,
>> what output do you get when you start molpro with "molpro -v" ?
>> You could also
>> turn on verbose output in mpirun.ch_gm as well (also giving a -v on
>> its command line).
>> best regards,
>> Kirk
>> On May 9, 2008, at 6:07 AM, Evgeniy Gromov wrote:
>>> Dear Kirk,
>>>
>>> Many thanks for your help! I made the settings
>>> in my CONFIG similar to yours and managed to compile
>>> the code. However now I am again in troubles as
>>> I cannot run any test. It looks like the executable
>>> is dead or cannot start as the output and errout
>>> files are empty. Did you have something like that
>>> when installing the code. Thanks!
>>>
>>> Best regards,
>>> Evgeniy
>>>
>>> Kirk Peterson wrote:
>>>> Dear Evgeniy,
>>>> I have a very similar cluster and in my case the various
>>>> (myrinet) CONFIG options are:
>>>> WRAPPER="/opt/mpich/myrinet/gnu/bin/mpirun.ch_gm"
>>>> MPI_LIB="/opt/mpich/myrinet/gnu/lib"
>>>> MYRINET_LIB="/opt/gm/lib"
>>>> MYRINET_LINKPARS="-lpthread -L/opt/mpich/myrinet/gnu/lib -lmpich -
>>>> L/opt/gm/lib -lgm"
>>>> LIBDIR_myrinet=/home/kipeters/ga-4-0/lib/LINUX
>>>> WRAPPER_myrinet=/opt/mpich/myrinet/gnu/bin/mpirun.ch_gm
>>>> of course the location of mpirun.ch_gm (/opt/mpich/myrinet/gnu/
>>>> bin in m case) should be in your path.
>>>> If this doesn't work for you, let me know and I can send my whole
>>>> CONFIG file to you.
>>>> regards,
>>>> Kirk
>>>> On May 8, 2008, at 6:13 AM, Evgeniy Gromov wrote:
>>>>> Dear molpro aficionados!
>>>>>
>>>>> I am attempting to install Molpro2006.1 in a parallel
>>>>> mode on a athlon cluster with a Myrinet network. I have
>>>>> installed GA with the GM Myricom library. When running
>>>>> configure, it prompts to choose among tcgmsg, mpi or myrinet.
>>>>> It is said in the manual that if myrinet network is available
>>>>> "myrinet" should be choosen. My question is what should be set
>>>>> then in the CONFIG file for: WRAPPER=, LIBDIR_myrinet= and
>>>>> WRAPPER_myrinet= ?
>>>>>
>>>>> On the other hand if I choose tcgmsg then:
>>>>>
>>>>> WRAPPER=.../ga-4-0-1/tcgmsg/ipcv4.0/parallel
>>>>> LIBDIR_tcgmsg=/home/thor2/egromov/prog/ga-4-0-1/lib/LINUX
>>>>> WRAPPER_tcgmsg=/usr/local/ga-4-0/tcgmsg/ipcv4.0/parallel
>>>>>
>>>>> Is this right?
>>>>> But when compiling I immediately get an error message from make:
>>>>>
>>>>> GNUmakefile:56: *** Cannot find 'parallel' command in PATH. Stop.
>>>>>
>>>>> I will very appreciate if someone could clarify (help to
>>>>> resolve) these
>>>>> two issues!
>>>>>
>>>>> Best wishes,
>>>>> Evgeniy
>>>>>
>>>>> --_______________________________________
>>>>> Dr. Evgeniy Gromov
>>>>> Theoretische Chemie
>>>>> Physikalisch-Chemisches Institut
>>>>> Im Neuenheimer Feld 229
>>>>> D-69120 Heidelberg
>>>>> Germany
>>>>>
>>>>> Telefon: +49/(0)6221/545263
>>>>> Fax: +49/(0)6221/545221
>>>>> E-mail: evgeniy at pci.uni-heidelberg.de
>>>>> _______________________________________
>>>>>
>>>>>
>>>>>
>>>
>>>
>>> --_______________________________________
>>> Dr. Evgeniy Gromov
>>> Theoretische Chemie
>>> Physikalisch-Chemisches Institut
>>> Im Neuenheimer Feld 229
>>> D-69120 Heidelberg
>>> Germany
>>>
>>> Telefon: +49/(0)6221/545263
>>> Fax: +49/(0)6221/545221
>>> E-mail: evgeniy at pci.uni-heidelberg.de
>>> _______________________________________
>>>
>>>
>>>
>
>
> --
> _______________________________________
> Dr. Evgeniy Gromov
> Theoretische Chemie
> Physikalisch-Chemisches Institut
> Im Neuenheimer Feld 229
> D-69120 Heidelberg
> Germany
>
> Telefon: +49/(0)6221/545263
> Fax: +49/(0)6221/545221
> E-mail: evgeniy at pci.uni-heidelberg.de
> _______________________________________
>
>
>
More information about the Molpro-user
mailing list