MPI has run out of internal group entries.

arras · Post by **arras** » Thu Aug 19, 2010 4:15 pm

Hi,
I'm trying to perform a parallel calculation on 4 procs using the syntax:
abinit -n 4 < ab.files > log

And I get the following message:

MPI has run out of internal group entries.
Please set the environment variable MPI_GROUP_MAX for additional space.
The current value of MPI_GROUP_MAX is 32
MPI has run out of internal group entries.
Please set the environment variable MPI_GROUP_MAX for additional space.
The current value of MPI_GROUP_MAX is 32
MPI: MPI_COMM_WORLD rank 0 has terminated without calling MPI_Finalize()
MPI: aborting job

There are 20 kpoints in my calculation and paral_kgb=0. Yet other calculations with less kpoints have passed allright.
Thus I confess I don't quite understand the complain...
I'm using Abinit 6.0.3 on SGI Altix 4700 (http://www.lrz.de/services/compute/hlrb/)

Thank you for any help..

Emmanuel ARRAS

mcote · Post by **mcote** » Mon Aug 23, 2010 3:37 pm

Hi,

We had similar problems on our SGI machine some times ago. We had to set the environment variables (in bash):

export MPI_GROUP_MAX=20000
export MPI_COMM_MAX=1000

We set these variables in the submission script, PBS in our case. I know that ABINIT uses one group(maybe more) per k-point. Somehow, the default for these environment variables in the SGI implementation is lower than on other machines.

Michel

arras · Post by **arras** » Wed Aug 25, 2010 12:15 pm

All right, I'll try that.
Thank you for the tip !

Emmanuel

arras · Post by **arras** » Fri Dec 03, 2010 12:25 pm

Hi again,
I'm using paral_kgb parallelisation with the following parameters:

paral_kgb 1
nband 448

npband 16
npfft 1
npkpt 4

acell 19.537533 28.2 12 Angstrom

on 64 processors with Abinit 6.4.1. The calculations runs allright, but the writting of the WFK file fails. The complain is :

Please set the environment variable MPI_TYPE_MAX for additional space.
The current value of MPI_TYPE_MAX is 8192

So I increased MPI_TYPE_MAX up to 1048576 ! And it still complains (yet on less lines). Is that normal ?? Is there a limit I should not cross on the number MPI_TYPE_MAX ? Or can I increase it until it works ?

Thanks !

Emmanuel

mcote · Post by **mcote** » Wed Dec 15, 2010 3:45 pm

I never had this problem but I presume that is because you are using MPI IO? I searched the web and it seems to be a specific SGI problem, for example see:

linux.math.tifr.res.in/programming-doc/lam/lam-users-guide.ps

and search for "for additional space".

It looks like you are doing the right thing to increase MPI_TYPE_MAX. Maybe you should increase it more...

Michel

ABINIT Discussion Forums

MPI has run out of internal group entries.

MPI has run out of internal group entries.

Re: MPI has run out of internal group entries.

Re: MPI has run out of internal group entries.

Re: MPI has run out of internal group entries.

Re: MPI has run out of internal group entries.