Sign in
Microsoft.com
 
Microsoft
 
 
 

locked
mpiexec fails to quit after ending of an invoked application RRS feed

 > 

  • Question

  • Question
    Sign in to vote
    0
    Sign in to vote

    Dear Sirs,

    I am aware and thankful that you are doing your best to make Microsoft MPI worth of its predecessors such as MPICH/MPICH2. Therefore I would be glad if you could help me to get rid of another annoying issue.

    In fact, I am using for years the Firefly quantum chemistry app, which also uses MS-MPI among others for parallel runs. However,  MS mpiexec always fails to quit itself after normal ending of the invoked program. This is not only an aesthetic matter, since because of it, scripts/batches/programs which drive Firefly through MS-MPI cannot be used efficiently. The developer team of Firefly is fully aware of this issue, asserting that "[...] this is a known bug of MS MPI version 7.1 and above with no known fix at Firefly level.  Namely, MS MPI fails to properly detect that Firefly job has terminated" and has suggested me to turn directly to Microsoft or as an alternative, to downgrade to MS-MPI v5, which allegedly does not have this issue.

    Please consider me at your disposal for each and every kind of tests you need to understand and fix the bug.

    Best regards,

    Maurizio


    • Edited by RotAnal Sunday, September 17, 2017 10:12 AM
    Sunday, September 17, 2017 10:09 AM

All replies

  • Question
    Sign in to vote
    0
    Sign in to vote

    Hi Maurizio,

    I am not familiar with Firefly. Is it freely available or do I need a license? If it is freely available or have some sort of trial license, I can try to reproduce the issues with your assistance and see what is causing mpiexec to not detect application termination. Does it still happen with MS-MPI v8.1 ?

    Anh

    Monday, September 18, 2017 12:16 AM
  • Question
    Sign in to vote
    0
    Sign in to vote

    Dear Ahn,

    thanks for your interest to the trouble that I exposed!

    As regards Firefly, it is freely available to researchers but it requires a free licence. In general, interested people have to fill and send a sheet to the development coordinator, namely prof. Alex A. Granovsky of University of Moscow. I think that in your particular case, you could get a trial licence if you explain him your aims. Alternatively, you could tell me what to do to test myself the mechanism of MPI. And yes, even 8.1 version has that failure.

    Maurizio

    Monday, September 18, 2017 8:06 PM
  • Question
    Sign in to vote
    0
    Sign in to vote

    Hi Maurizio,

    I'll get in touch with Prof. Granovsky to investigate this issue. I will update this thread when we have more information

    Anh

    Saturday, September 23, 2017 6:34 PM
  • Question
    Sign in to vote
    0
    Sign in to vote

    Dear Anh,

    did you eventually succeed in contacting Prof. Granovsky in order to arrange to solve the trouble? I ask you also on behalf of him, who told me that the Firefly web site due to technical reasons had been down for few weeks but now it is fine. Therefore if you tried to reach him with no success, I would advise you to try again.

    Best regards

    Maurizio

    Friday, October 27, 2017 5:06 PM