[Wien] lapw2para very slow

Peter Blaha pblaha at theochem.tuwien.ac.at
Sun Jul 23 11:35:04 CEST 2006


Why are you using    mpiexec  ??? I don't know what mpiexec does, but
Usually   mpiXXX scripts start the job   N-times, ant that's for sure not 
what you want.

For the DOS:
Either run AFTER the scf cycle one   x lapw1 -p   step WITHOUT SCRATCH
(I'd do this, because for a good DOS you need more k-points anyway),
or modify your script in a "clever" way, so that after the scf cycle (but 
within the pbs-job) you cp the vector_* files from the different node-tmp 
directories to your global working dir. Something like:

ssh nodeX "cp /tmp/*vector_* $PBS_O_WORKDIR"

You can use this also to remove the scratch files afterwards to clean the 
scratch dir.



> Dear Prof. Blaha,
> thank you very much for your response. 
>  
> First of all, the CPU time is normal but the elapsed time 
> is very big. I'll discuss this with the sys admin.
>  
> In the meantime, I would like to try 
> your suggestion. How do I define the local SCRATCH VARIABLE on  
> each node. If a calculation is done and I have case.vector 
> files the SCRATCH directory on individual nodes, how do 
> I retrieve them for the DOS calculation, for example. 
> Is it possible to include your suggestion to the pbs 
> job script. Please find below my pbs script for an 18 k-point job.
>  
> =========================================
>  
> #!/bin/bash
> #PBS -l walltime=3400:00:00
> #PBS -l nodes=3:ppn=2
> #PBS -o testrun.out
> #PBS -e testrun.err
> #PBS -q default
> #PBS -N bulk
> #PBS -j oe
> 
> cd $PBS_O_WORKDIR
>  
> # Find free nodes
> for NODE in `cat $PBS_NODEFILE | sort -u`; do
>         ssh $NODE /opt/mpich/p4-intel/sbin/cleanipcs
> done
>  
> #create .machines file with 3 k-points per processor 
> rm -f .machines
> awk '{print "3:"$1}' $PBS_NODEFILE > .machines
> echo 'granularity:1' >>.machines
> 
> /opt/mpiexec/bin/mpiexec runsp_lapw -p -so -i 100 -ec 0.000001
>  
> ===============================================================
> 


                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------


More information about the Wien mailing list