[mpich-discuss] mpiexec and working directory

[Reuti]

Hi,

Am 07.11.2008 um 20:35 schrieb Tiago Silva:

> Problem solved. I remembered that there was also an mpd version of  
> mpich2 on the same machine and it works as expected:
>
> [tsilva at admin1 examples]$ mpd &
> [1] 12297
> [tsilva at admin1 examples]$ /n1ge6/mpich2-1.0.3/bin/mpiexec -n 2 pwd
> /n1ge6/mpich2-1.0.3/examples
> /n1ge6/mpich2-1.0.3/examples
>
> I wasn't using it because I couldn't  integrate it with SGE, unlike  
> the smpd running with -nopm. But that is another story.

I will update the Howto at http://gridengine.sunsource.net/howto/ 
mpich2-integration/mpich2-integration.html in the next days to also  
include a Tight Integration for the MPD method for startup.

-- Reuti

>
> Thanks for the help,
> Tiago
>
>
> Jayesh Krishna wrote:
>>
>>  Hi,
>>   When you install MPICH2 make sure that you install the latest  
>> stable version, 1.0.8. You could also ask your sysadmin to install  
>> MPICH2+smpd in another directory  so that you can test MPICH2+smpd  
>> (1.0.8).
>>
>> Regards,
>> Jayesh
>>
>> -----Original Message-----
>> From: mpich-discuss-bounces at mcs.anl.gov [mailto:mpich-discuss- 
>> bounces at mcs.anl.gov] On Behalf Of Tiago Silva
>> Sent: Friday, November 07, 2008 12:26 PM
>> To: mpich-discuss at mcs.anl.gov
>> Subject: Re: [mpich-discuss] mpiexec and working directory
>>
>>
>> So could that explain this behavior? I need to justify asking a  
>> very busy sysadmin to get the daemons  on a 25 node cluster.
>>
>> Tiago
>>
>> Rajeev Thakur wrote:
>> > Oh, you are using SMPD. Can you try with the default MPD process  
>> manager?
>> > You have to build MPICH2 again for that.
>> >
>> > Rajeev
>> >
>> >  >> -----Original Message-----
>> >> From: mpich-discuss-bounces at mcs.anl.gov
>> >> [mailto:mpich-discuss-bounces at mcs.anl.gov] On Behalf Of Tiago  
>> Silva
>> >> Sent: Friday, November 07, 2008 12:18 PM
>> >> To: mpich-discuss at mcs.anl.gov
>> >> Subject: Re: [mpich-discuss] mpiexec and working directory
>> >>
>> >> Thanks Rajeev, this shows the problem nicely:
>> >>
>> >> [tsilva at admin1 work]$ /n1ge6/mpich2/smpd/bin/mpiexec -nopm -ssh  
>> -l -n
>> >> 2 pwd /raid1/tsilva /raid1/tsilva
>> >> [tsilva at admin1 work]$ /n1ge6/mpich2/smpd/bin/mpiexec -nopm -ssh -l
>> >> -wdir /tmp/ -n 2 pwd /raid1/tsilva /raid1/tsilva
>> >> [tsilva at admin1 work]$ ls runjob.err
>> >> runjob.err
>> >> [tsilva at admin1 work]$ /n1ge6/mpich2/smpd/bin/mpiexec -nopm -ssh -l
>> >> -wdir /tmp/ -n 2 ls -l runjob.err
>> >> ls: runjob.err: No such file or directory
>> >> ls: runjob.err: No such file or directory
>> >>
>> >> Could this have anything to do with running mpiexec deamonless
>> >> (although mpich was compiled with smpd)?
>> >> I noticed the same behavior with my regular setup of also  
>> deamonless
>> >> mpirun and mpich1. Only after calling mpi_init in the code did the
>> >> working dir change from being $home to the local dir.
>> >>
>> >> Cheers,
>> >> Tiago
>> >>
>> >> Rajeev Thakur wrote:
>> >>    >>> Can you try these simple commands:
>> >>>
>> >>> mpiexec -l -n 2 pwd   >>>   ## shows if you execute in the dir  
>> you think you are in
>> >>>
>> >>> mpiexec -l -wdir /tmp -n 2 pwd   >>>   ## similar to above but  
>> shows you moved to /tmp
>> >>>
>> >>> mpiexec -l -n 2 ls -l some_filename   >>>   ## shows if some  
>> expected file really is in the working dir
>> >>>
>> >>> >>>
>> >>>  >>>      >>>> -----Original Message-----
>> >>>> From: mpich-discuss-bounces at mcs.anl.gov
>> >>>> [mailto:mpich-discuss-bounces at mcs.anl.gov] On Behalf Of
>> >>>> tsilva at coas.oregonstate.edu
>> >>>> Sent: Friday, November 07, 2008 1:26 AM
>> >>>> To: mpich-discuss at mcs.anl.gov
>> >>>> Subject: Re: [mpich-discuss] mpiexec and working directory
>> >>>>
>> >>>> Quoting Rajeev Thakur <thakur at mcs.anl.gov>:
>> >>>>
>> >>>>    >>>>        >>>>> If you cd to the directory containing  
>> the executable, say
>> >>>>>      >>>>>          >>>> a.out, and
>> >>>>    >>>>        >>>>> do "mpiexec -n 2 a.out" does it expect  
>> to find input
>> >>>>>          >> files in your
>> >>    >>>>> home directory? That's odd.
>> >>>>>      >>>>>          >>>> That is correct. I have tried  
>> another simpler code and the same
>> >>>> thing happens.
>> >>>>
>> >>>>    >>>>        >>>>>> -----Original Message-----
>> >>>>>> From: mpich-discuss-bounces at mcs.anl.gov
>> >>>>>> [mailto:mpich-discuss-bounces at mcs.anl.gov] On Behalf Of
>> >>>>>>            >> Tiago Silva
>> >>    >>>>>> Sent: Thursday, November 06, 2008 6:02 PM
>> >>>>>> To: mpich-discuss at mcs.anl.gov
>> >>>>>> Subject: Re: [mpich-discuss] mpiexec and working directory
>> >>>>>>
>> >>>>>> Rajeev Thakur wrote:
>> >>>>>>        >>>>>>            >>>>>>> 1.0.3 is an ancient  
>> version of MPICH2. Try the latest
>> >>>>>>>          >>>>>>>              >>>>>> version, 1.0.8.
>> >>>>>>        >>>>>> I will ask my systems administrator to do so.
>> >>>>>>
>> >>>>>>        >>>>>>            >>>>>>> The parameters to mpiexec  
>> are not identical between MPICH-1 and
>> >>>>>>> MPICH2. Do mpiexec --help. Make sure you are calling
>> >>>>>>>          >>>>>>>              >>>>>> MPICH2's mpiexec, not  
>> MPICH-1's.
>> >>>>>>        >>>>>> I am actually invoking mpiexec with its full  
>> path. Is there any
>> >>>>>> incorrect parameter on my mpiexec call?
>> >>>>>> I
>> >>>>>>        >>>>>>            >>>>>>> If your input files are in  
>> the same directory as the
>> >>>>>>>          >>>>>>>              >>>>>> executable, you
>> >>>>>>        >>>>>>            >>>>>>> should not need to specify  
>> anything.
>> >>>>>>>
>> >>>>>>>          >>>>>>>              >>>>>> But that is exactly  
>> the problem. The executable can't find
>> >>>>>>        >>>>>>            >>>> the input
>> >>>>    >>>>        >>>>>> files on the working directory.
>> >>>>>>
>> >>>>>>
>> >>>>>>        >>>>>>            >>>>>      >>>>>          >>>>
>> >>>>    >>>>        >>>  >>>      >>    >
>> >
>