[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [ESPResSo] Espresso over infiniband
From: |
Tristan Bereau |
Subject: |
Re: [ESPResSo] Espresso over infiniband |
Date: |
Wed, 10 Dec 2008 10:26:04 -0500 |
User-agent: |
Thunderbird 2.0.0.6 (X11/20070801) |
Dear Olaf,
Thanks for your help. Indeed the mbtools package was writing to a
compressed file (via the checkpoint_set command). Axel explained to me
this procedure invoked a call to the fork() command, which is not
supported on Infiniband.
I guess I won't compress files !
Best,
Tristan
Olaf Lenz wrote:
> Hi!
>
> Tristan Bereau wrote:
>> However, when running Espresso, my job crashes after a while when trying
>> to write to a blockfile (one of the Espresso blockfile write command)
>> because of a "broken pipe." I'm a bit puzzled because this does not
>> necessarily happen at the first call of the function. The job might be
>> able to write a few blockfiles before crashing. And, the job *only*
>> crashes because of a "broken pipe."
>> Do I have some stability issues with the connection ?
>
> I do think that the problem is not directly one of ESPResSo. Most
> probably, the problem is connected to your MPI implementation.
> I know that OpenMPI seems to have a strange usage of different Unix
> signals, and therefore is making some problems with opening and
> closing files, and in particular when you are using pipes to external
> programs.
>
> Are you writing out to a compressed file? In that case, the first
> thing you could try is not to write to a compressed file, but to a
> normal file instead.
> Otherwise, are you reading in another file in your script? If so, keep
> care that the file is completely read, before you close it. Also, do
> not forget to explicitly close the file.
>
> If none of this helps, it would be helpful if you could send around
> the script that creates the problems, ideally reduced to the lines of
> the script that actually fail.
>
> Best regards
> Olaf
>