The ABINIT Users Mailing List ( CLOSED )

[Latévi Max LAWSON DAKU <max.lawson@unige.ch>]

Dear Savas,

Many thanks for your answer and your tips.

I have adapted your script to my PBS+MPI
environment; the job is queued.

Concerning the serial calculations, the jobs
are effectively spawned to the other nodes.

All the best,
Max

P.S. Note that this was also our system manager
who points me at the problem with the parallel
job !-)



Savas Berber wrote:

> Dear Max,
>
> Here is the script I used on PBS job submitting system
>
> #!/bin/sh
> #@$-q pilim
> #@$-lP 4
> #@$-o error.log
> #@$-eo
> #
> cd $QSUB_WORKDIR
> scrun -nodes=4,stat=detail scatter -node 0 == ../abinip < test1.files
> > test1.log
>
> You are passing most of the arguments in command line. Your way of
> submitting should be OK. The system you are using may be well
> configured, but our system seems to be picky about many things.
> Compiling on your PC cluster may be a good idea.
>
> By the way, how do you submit serial jobs? First, I was submitting like
>
> ../abinis < test1.files > test1.log
>
> in the job submission script, then I received an angry email from
> system manager. Apparently, my serial jobs were running on the master
> node that controls the whole cluster and PBS. They were never spawned
> to other nodes. He told me to use
>
> scrun -nodes=1,stat=detail scatter == system ../abinis < test1.files >
> test1.log
>
> which starts the system on remote node, and runs serial job there.
>
> Good Luck!
>
> --Savas
>
> Latévi Max LAWSON DAKU wrote:
>
>> Dear Savas,
>>
>> Many thanks for your answer.
>>
>> I am not using SCORE.
>>
>> Jobs are submitted through the PBS system, as follows
>>
>>     $ qsub -l nodes=4 -q pilim myjob.sh
>>
>> and it is up to PBS to correctly set up the parameters
>> required for  the parallel calculations to proceed. The
>> script myjob.sh reads like
>>
>>    #!/bin/bash
>>
>>    mpirun ~ABINIT/abinip < ab.in > ab.out
>>
>> Obviously everything is fine, at the begining. But all the
>> job is done by the master process:
>>
>> --------------------------------------------------------------------------------
>>
>>
>> P newkpt: treating      9 bands with npw=    5026 for ikpt=   1 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5037 for ikpt=   2 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5037 for ikpt=   3 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5043 for ikpt=   4 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5054 for ikpt=   5 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5041 for ikpt=   6 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5034 for ikpt=   7 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5031 for ikpt=   8 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5018 for ikpt=   9 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5035 for ikpt=  10 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5054 for ikpt=  11 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5052 for ikpt=  12 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5056 for ikpt=  13 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5049 for ikpt=  14 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5028 for ikpt=  15 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5037 for ikpt=  16 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5049 for ikpt=  17 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5056 for ikpt=  18 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5060 for ikpt=  19 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5048 for ikpt=  20 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5045 for ikpt=  21 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5050 for ikpt=  22 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5048 for ikpt=  23 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5058 for ikpt=  24 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5061 for ikpt=  25 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5053 for ikpt=  26 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5046 for ikpt=  27 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5052 for ikpt=  28 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5058 for ikpt=  29 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5043 for ikpt=  30 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5043 for ikpt=  31 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5025 for ikpt=  32 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5032 for ikpt=  33 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5040 for ikpt=  34 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5052 for ikpt=  35 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5060 for ikpt=  36 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5056 for ikpt=  37 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5048 for ikpt=  38 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5046 for ikpt=  39 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5053 for ikpt=  40 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5059 for ikpt=  41 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5056 for ikpt=  42 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5065 for ikpt=  43 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5058 for ikpt=  44 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5060 for ikpt=  45 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5064 for ikpt=  46 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5055 for ikpt=  47 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5046 for ikpt=  48 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5058 for ikpt=  49 by
>> node    0
>> P newkpt: treating      9 bands with npw=    5046 for ikpt=  50 by
>> node    0
>>   newkpt: prtvol=0 or 1, do not print more k-points.
>>  
>>  setup2: Arith. and geom. avg. npw (full set) are    5050.535   
>> 5050.525
>>
>> =========================================================
>>
>>
>> I going to try to compile the source on the cluster.
>>
>> All the best,
>> Max
>>
>>
>>
>>
>> Savas Berber wrote:
>>
>>
>>> Are you using SCORE to spawn your jobs?
>>>
>>> In any case, you have to use mpirun to start your job.
>>>
>>> mpirun -np 4 abinip
>>>
>>> to run on 4 processors. The binary you donwnloaded may be for a
>>> different implementation.
>>>
>>> Good luck!
>>>
>>> --Savas
>>>
>>> Latévi Max LAWSON DAKU wrote:
>>>
>>>
>>>> Dear All,
>>>>
>>>> I am trying to use ABINIT (v4.6.5) in parallel mode on a myri
>>>> cluster with MPI. But there is something going wrong.
>>>> At startup, the processes are correctly created on the
>>>> different nodes but it seems that the real calculations are
>>>> being made by the master process, only; the secondary
>>>> processes being idle...
>>>>
>>>> Did I miss something ?
>>>>
>>>> I did not put any special instructions in the input file, (which
>>>> works well for serial calculations), so as to benifit from the
>>>> automatic parallelism on the k-point (is it correct); and the
>>>> binary I am using have been downloaded from the ABINT
>>>> dowload page.
>>>>
>>>> I thank you in advance for any clue you could give me.
>>>>
>>>> All the best,
>>>> Max
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>
>>>
>>
>
>

-- 
***********************************************
Latevi Max LAWSON DAKU
Departement de chimie physique 
Universite de Geneve - Sciences II
30, quai Ernest-Ansermet  
CH-1211 Geneve 4 
Switzerland 

Tel: (41) 22/379 6548 ++ Fax: (41) 22/379 6103
***********************************************