I_MPI_DEBUG output

I_MPI_DEBUG output

I have a client that has a problem with my MPI enabled product.
To find out more of his problem i have asked him to run the following command

mpiexec -n 2 -localonly -genv I_MPI_DEBUG 100 program

I get the following output and I wonder why its [-1], on other pc's its [#rank] and why no hostname is specified. After the last line i receive an Appcrash, this is directly after MPI_Init.

Any idea how to attack this problem?

[-1] MPI startup(): I_MPI_DEBUG=100

[-1] MPI startup(): set domain to {0,1,2,3,4,5,6,7} on node

[-1] MPI startup(): Recognition level=1. Platform code=3. Device=0

[-1] MPI startup(): Parent configuration:(intra=0 inter=0 flags=0), (code=3 ppn=0)

.

.

.

[-1] Rank Pid Node name Pin cpu

[-1] 0 3684 {0,1,2,3,4,5,6,7}

Thanks for your input and help

Jesper

4 posts / 0 nouveau(x)
Dernière contribution
Reportez-vous à notre Notice d'optimisation pour plus d'informations sur les choix et l'optimisation des performances dans les produits logiciels Intel.

Hi Jesper,

Could you please provide information of the Intel MPI version? If your customer sees this error - can you reproduce the issue on your cluster with the same command line? Does the application crash without I_MPI_DEBUG? Could you suggest running the application without '-localonly'?
At the moment it's not clear the reason of such behaviour.

Regards!
Dmitry

Hi Dmitry and thanks for you reply

The application crashes in the same way without the I_MPI_DEBUG andI have instructed our customer to try and do a run with -host instead of -localonly. I am currently waiting for a reply.
Sadly the problem cannot be reproduced on our cluster, however i am trying to get direct access to his computer (the customer has the same problem on several computers - all with a Hungarian windows 7)
The mpiexec -version output is:

Intel MPI Library for Windows* OS, Version 4.0 Build 2/18/2010 1:00:47 PM

Copyright (C) 2007-2010, Intel Corporation. All rights reserved.

Which indicate the correct version - we install the 4.0.0.012 MPI-RT package.

In the output from I_MPI_DEBUG as i read it the number in the [ ] should be the rank of the process right?
and do you know of a correct example where the hostname is not listed in the last part of the output?

Regards
Jesper

Jesper,

-1 in square brackets means that the rank number was not assigned yet. Usually debug output goes from rank 0, but if initialization was not complete properly then you can see [-1]. It happens somethimes but not so often.

>do you know of a correct example where the hostname is not listed in the last part of the output
No, I don't. Does the customer use pathes with local characters (non-english)? Might be this the reason?

Regards!
Dmitry

Laisser un commentaire

Veuillez ouvrir une session pour ajouter un commentaire. Pas encore membre ? Rejoignez-nous dès aujourd’hui