[gmx-users] Parallel running problem

Erik Lindahl lindahl at sbc.su.se
Fri Sep 16 15:47:56 CEST 2005


Hi,

This is an problem with your MPI installation.

The installation is broken/misconfigured, or you have your PATH  
settings mixed up. Ask your sysadm or check the documentation for the  
MPI package you are using.

The configure script checks for the normal MPI cc scripts (and finds  
mpicc).

However, when it tries to compile a small test program that calls  
MPI_Init() it doesn't work, even if it tries to link with -lmpi  
explicitly.

This is essentially what the script is testing to build:

-----
#include <mpi.h>

int
main(int argc,char **argv)
{
     MPI_Init(&argc,&argv);
     exit 0;
}
----

Cheers,

Erik


On Sep 16, 2005, at 2:44 PM, Alan Dodd wrote:

> For completeness, this is the configure command I'm
> using for installing fftw:
> ./configure --prefix=/home/ad0303/fftw --enable-float
> --enable-type-prefix --enable-mpi
> And this is the result:
>
> checking for mpicc... mpicc
> checking for MPI_Init... no
> checking for MPI_Init in -lmpi... no
> configure: error: couldn't find mpi library for
> --enable-mpi
>
> I'm certain there's some critical command I'm not
> specifying here - what did you mean by "I also linked
> FFTW against MPICH"?
>
> --- "Peter C. Lai" <sirmoo at cowbert.2y.net> wrote:
>
>
>> On Fri, Sep 16, 2005 at 03:23:07AM -0700, Alan Dodd
>> wrote:
>>
>>> Thanks for all your help, I thought I had compiled
>>> with MPI but from trying to reinstall, it appears
>>>
>> not.
>>
>>> The system I'm trying to install on is using
>>>
>> mpich,
>>
>>> rather than lammpi.  I wouldn't have thought this
>>> would be a problem, but installing fftw locally
>>> doesn't work - it can't find the mpi libraries.
>>>
>> Both
>>
>>> using the rpms and compiling from source seem to
>>> produce similar errors.  I'm pretty sure others
>>>
>> have
>>
>>> used mpich, so have any of you come across a
>>>
>> similar
>>
>>> problem (and, ideally, a solution?)
>>> Thanks,
>>> Alan
>>>
>>>
>>
>> I dunno, I recently (a few months ago, really)
>> compiled 3.2.1 against
>> mpich1.2.5.2 initially and then against mpich1.2.6
>> on a dual-socket p3
>> with gcc3.4.2 and ran all the test suites with no
>> problems (I was actually
>> running gromacs-mpich to debug the new FreeBSD
>> kernel scheduler while getting
>> some side work out of the otherwise idle cpus), so I
>> really don't know what
>> your problem is either [i.e. WORKSFORME albeit on a
>> different platform] :(
>>
>> Note that I also linked FFTW against MPICH - I think
>> this is a critical step
>> (and everything was built as single precision, but I
>> vaguely remember running
>> double over mpich without any problems either).
>>
>>
>>> --- David van der Spoel <spoel at xray.bmc.uu.se>
>>>
>> wrote:
>>
>>>
>>>
>>>> On Wed, 2005-09-07 at 05:18 -0700, Alan Dodd
>>>>
>> wrote:
>>
>>>>> Hello Gromacs users,
>>>>> I gather this problem is similar to many
>>>>>
>> previous,
>>
>>>> but
>>>>
>>>>> can't see an obvious solution in the replies
>>>>>
>> to
>>
>>>> any of
>>>>
>>>>> them.  I've been trying to get GROMACS to run
>>>>>
>> on
>>
>>>> this
>>>>
>>>>> sample dual-core, dual-socket opteron box that
>>>>>
>> we
>>
>>>> have
>>>>
>>>>> on loan.  Despite my best efforts, I seem
>>>>>
>> unable
>>
>>>> to
>>>>
>>>>> get mdrun to understand that it's supposed to
>>>>>
>> run
>>
>>>> on
>>>>
>>>>> more than one node.  I'm telling it to do so,
>>>>>
>> and
>>
>>>> it
>>>>
>>>>> even appreciates it's supposed to in the
>>>>>
>> output
>>
>>>> (see
>>>>
>>>>> below), but then decides I've told it to run
>>>>>
>> on
>>
>>>> just
>>>>
>>>>> the one and dies.  Has anyone any idea what's
>>>>>
>>>> going
>>>>
>>>>> wrong?  Is it just some kind of
>>>>>
>> incompatibility
>>
>>>> with
>>>>
>>>>> mpich/the hardware?
>>>>>
>>>> Have you compiled with MPI?
>>>>
>>>> you can check by typing
>>>> ldd `which mdrun`
>>>> It should show some MPI libraries.
>>>>
>>>> Dual core opterons run fine by the way. We have
>>>>
>> a
>>
>>>> brand new cluster
>>>> humming along at 85 decibel.
>>>>
>>>>>
>>>>> Input:
>>>>> mpirun -np 4 -machinefile machines mdrun -np 4
>>>>>
>>
>>
>>>>>
>>>>> mdrun output:
>>>>> for all file options
>>>>>          -np    int      4  Number of nodes,
>>>>>
>> must
>>
>>>> be
>>>>
>>>>> the same as used for
>>>>>                             grompp
>>>>>          -nt    int      1  Number of threads
>>>>>
>> to
>>
>>>> start
>>>>
>>>>> on each node
>>>>>       -[no]v   bool     no  Be loud and noisy
>>>>> -[no]compact   bool    yes  Write a compact
>>>>>
>> log
>>
>>>> file
>>>>
>>>>>   -[no]multi   bool     no  Do multiple
>>>>>
>>>> simulations in
>>>>
>>>>> parallel (only with
>>>>>                             -np > 1)
>>>>>    -[no]glas   bool     no  Do glass
>>>>>
>> simulation
>>
>>>> with
>>>>
>>>>> special long range
>>>>>                             corrections
>>>>>  -[no]ionize   bool     no  Do a simulation
>>>>>
>>>> including
>>>>
>>>>> the effect of an X-Ray
>>>>>                             bombardment on
>>>>>
>> your
>>
>>>> system
>>>>
>>>>>
>>>>>
>>>>> Back Off! I just backed up md.log to
>>>>>
>> ./#md.log.5#
>>
>>>>> Reading file short.tpr, VERSION 3.2.1 (single
>>>>> precision)
>>>>> Fatal error: run input file short.tpr was made
>>>>>
>> for
>>
>>>> 4
>>>>
>>>>> nodes,
>>>>>              while mdrun expected it to be for
>>>>>
>> 1
>>
>>>>> nodes.
>>>>>
>>>>>
>>>>> Alan Dodd (University of Bristol)
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>> _
>>>> _______________________________________________
>>>> gmx-users mailing list
>>>> gmx-users at gromacs.org
>>>>
>>>>
>> http://www.gromacs.org/mailman/listinfo/gmx-users
>>
>>>> Please don't post (un)subscribe requests to the
>>>> list. Use the
>>>> www interface or send it to
>>>> gmx-users-request at gromacs.org.
>>>>
>>>>
>>>
>>>
>>>
>>>
>>> __________________________________
>>> Yahoo! Mail - PC Magazine Editors' Choice 2005
>>> http://mail.yahoo.com
>>> _______________________________________________
>>> gmx-users mailing list
>>> gmx-users at gromacs.org
>>> http://www.gromacs.org/mailman/listinfo/gmx-users
>>> Please don't post (un)subscribe requests to the
>>>
>> list. Use the
>>
>>> www interface or send it to
>>>
>> gmx-users-request at gromacs.org.
>>
>> -- 
>> Peter C. Lai
>> Cesium Hyperfine Enterprises
>> http://cowbert.2y.net/
>>
>>
>>
>
>
>
>
> __________________________________
> Yahoo! Mail - PC Magazine Editors' Choice 2005
> http://mail.yahoo.com
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
>
>




More information about the gromacs.org_gmx-users mailing list