The ABINIT Users Mailing List ( CLOSED )

[Alessandro Fortunelli <fortunelli@ipcf.cnr.it>]

Dear Matthieu,

I tried all of your suggestions, but I did not find a way to solve the
problem: I confirmed that my jobs do not run in parallel, but I do not
understand why. It's possibly something connected with mpich, so I was
looking for somebody that had already experienced my same problems.

Thanks anyway for your suggestions,

Alessandro


In data 29/8/2008, "matthieu verstraete"
<matthieu.jean.verstraete@gmail.com> ha scritto:

>your number of kpoints is equal to mkmem, so your abinip is certainly not
>running on 2 processors...
>
>try with mynodes=
>growth1
>growth1
>
>for instance, or even better, using 2 machines so you can be sure both are
>being used
>
>Are you sure of the syntax for mpiexec? I usually use mpirun (which has the
>syntax you give), and there may be a difference...
>
>
>> >Do you get any message in your log or output file related to this problem
>> ?
>>
>> Well, as you can see from the attached log and output files, there are
>> indications that the system is using only one cpu (e.g., only node 0 is
>> mentioned), but I cannot understand wh
>
>the main log is only written to by the mother process. Check the other _LOG_
>files. Again if they are not there you are not running in parallel...
>
>
>>
>> ------------------------------------------------------------------------------
>>
>> I followed the run through top command for the 2' something of the run, and
>> I always saw ONLY ONE abinip running. The cpu time was identical with
>> abinis.
>
>In a short run, much of the time may be spent with only one processor
>running actively. You might try making the run a bit longer (increase ecut
>or nkpt or something). In top try "u" for user and then your user name, to
>see all of your processes (including inactive ones). You should have 2 or 4
>abinip-s (the doubling can happen on some platforms due to the way the mpi
>works, but the other 2 don't do anything they just sit there -anyone know
>why this happens?)
>
>good luck
>
>
>Matthieu
>
>

Alessandro Fortunelli
IPCF-CNR, via G. Moruzzi, 1
56124 - Pisa -Italy
e-mail: fortunelli@ipcf.cnr.it
tel. +39-050-3152447
fax  +39-050-3152442
cel. +39-349-2987108
Home-page: http://h2.ipcf.cnr.it/alex/af.html