The ABINIT Users Mailing List ( CLOSED )

[Michel Cote <Michel.Cote@umontreal.ca>]

We also have some problems with the phonon calculations and MPI. Abinit creates MPI_GROUP for the communications but does not free these groups and MPI runs out of groups and stop. This is certainly dependent on the implementation of the MPI. We have observed this problem on a SGI Altix machine, but it might be the same problem on your machine.

You could increase the numbers of groups in the MPI by setting the environment variable MPI_GROUP_MAX to a larger value than the default. Also, it might help to split your runs, i.e., one mpirun per dataset.

We are working to modify the code to fix this problem, but we have not yet implemented in the code.

In anycase, I am interested to know whether this fixes your problem.

Michel

Le 07-07-25 à 21:34, Y X WANG a écrit :

Dear Dr Matthieu,

Thank you for your reply!

I use command "nohup mpiexec -machinefile f -np 16

abinip <bto.files>log&". I did not use "nolocal".

mpi version is mpich2-1.0.5p4.

Thank you again.

WANG Yuanxu

--- Matthieu Verstraete <mjv500@york.ac.uk> wrote:

This looks like the same bug I had on Mac OSX. Try

mpirun without the 

-nolocal argument, that might help. More generally,

there is definitely a 

bug here, but I haven't been able to trace it.

Matthieu

-- 

================================================================

Dr. Matthieu Verstraete                

mailto:mjv500@york.ac.uk

Dept. of Physics, University of York,     tel: +44

1904 43 22 08

Heslington, YO10 5DD York, United Kingdom fax: +44

1904 43 22 14

____________________________________________________________________________________

Got a little couch potato? 

Check out fun summer activities for kids.

http://search.yahoo.com/search?fr=oni_on_mail&p=summer+activities+for+kids&cs=bz 

_____________________________________

Michel Cote

Departement de physique

Universite de Montreal

emai: Michel.Cote@umontreal.ca

web: http://www.phys.umontreal.ca/~michel_cote