[Wien] Trouble executing lapw1 (Cannot find libfftw.so.3)

[Laurence Marks]

Let me expand on my response. Much of what I say below can be found in
previous posts, but it seems that it is useful to repeat it every now
and again.

When you use mpi with Wien2k, you need to realize that you are moving
in to an area where how the sys_admin has setup the system matters,
and more often than you might think it has been done wrong. Wien2k is
the best code I know of for finding OS bugs. In addition, the default
way to implement mpi with PBS or similar is to assume that the user is
only going to run a single task, and Wien2k does much better than this
but breaks the mold.

All mpi flavors have issues, some severe, some not. Some I know of are:

a) Openmpi does not by default export environmental variables, I am
told for security reasons. You therefore have to force the appropriate
ones to be used. I am 99.9% confident this is the source of the
original issue in this thread.

b) Some earlier versions of openmpi are broken and don't work with Wien2k.

c) There are two bugs in ssh which can effect Wien2k operation.
* One is that killing the ssh task on the node that launches mpi (in
impi/mvapach for instance) may or may not kill the ssh task on the
child -- different versions of ssh are different.
* The second is that in some cases the process on the child can
inappropriately terminate leaving a zombie on the node that launched
the ssh. I know this can happen due to temporary hardware issues. As a
consequence the process will hang forever.

d) Some flavors of mpi do not follow the format for machines files
that others do, so need different treatment. Unlike Fortran
system-level codes are often not standard.

There is a small collection of notes/utilities for mpi in
http://www.wien2k.at/reg_user/unsupported/SRC_mpiutil.tar.gz .
Personally I use Machines2W all the time with PBS and other similar
systems as it removes the need to write specialized scripts. It knows
about a decent number of OS, and hopefully others will provide more
information on any others so it can be expanded. It contains some work
arounds for OS bugs I know of, but not everything.

On Thu, Jan 16, 2014 at 1:22 PM, Laurence Marks
<L-marks at northwestern.edu> wrote:
> By default openmpi does not export $LD_LIBRARY_PATH, so you need in
> parallel options something like
>
> setenv WIEN_MPIRUN "mpirun -x LD_LIBRARY_PATH -x PATH -np _NP_
> -machinefile _HOSTS_ _EXEC_"
>
>
> (See SRC_mpiutil which has some other examples)
>
> On Wed, Jan 15, 2014 at 4:23 PM, Joshua Davis
> <davis101 at chemistry.msu.edu> wrote:
>> Dear Wien Users
>>
>> I have Compiled Wien2K 13.1 on our institutions High Performance Computer
>> Cluster.
>> It was complied with ifort13.01,mkl10.3,FFTW3.3.3, and OpenMPI1.4.4.  It all
>> compiled perfectly with no errors.  However, when I go execute the program I
>> get the error:
>>
>> lapw1: error while loading shared libraries: libfftw3.so.3: cannot open
>> shared object file: No such file or directory
>>
>> my $LD_LIBRARY_PATH is:
>>
>> /opt/software/fftw/3.3.3--Intel-13.0.1.117--OpenMPI-1.4.4/lib:/opt/software/R/2.15.1--Intel-12.0.0.084/lib64/R/lib:/opt/software/boost/1.52.0--Intel-13.0.1.117--OpenMPI-1.4.4/lib:/opt/software/ClusterStudio/2013.1/composer_xe_2013.1.117
>>
>> Where the library should be is located int the first pathway listed.  I am
>> very confused as to why this I am getting this error when I feel the library
>> path is clearly stated.
>>
>> For other information, the HPCC that is being used is made up of different
>> machines with differing memory and cpus, however they are all linked by high
>> speed infiniband.
>>
>> Thank you for your help
>>
>> ----------------------------------------------------------------------------------------------------------------
>> Joshua D. Davis
>> davis.d.josh at gmail.com
>> Cell: (734)945-6031
>>
>> Graduate Assistant
>> Department of Chemistry
>> Michigan State University
>>
>> 578 S. Shaw Lane, room 432
>>
>> East Lansing, MI 48824
>> -----------------------------------------------------------------------------------------------------------------
>
>
>
> --
> Professor Laurence Marks
> Department of Materials Science and Engineering
> Northwestern University
> www.numis.northwestern.edu 1-847-491-3996
> "Research is to see what everybody else has seen, and to think what
> nobody else has thought"
> Albert Szent-Gyorgi



-- 
Professor Laurence Marks
Department of Materials Science and Engineering
Northwestern University
www.numis.northwestern.edu 1-847-491-3996
"Research is to see what everybody else has seen, and to think what
nobody else has thought"
Albert Szent-Gyorgi