[Pw_forum] Distributing phonon calculations todifferent machines

Dal Corso Andrea dalcorso at sissa.it
Wed Jul 7 14:06:54 CEST 2010


On Tue, 2010-07-06 at 17:55 +0800, Huiqun Zhou wrote:
> Thanks, Andrea.
> 
> As indicated in the script below, I have copied all files and directories 
> created
> by pw.x run.
> 
> if test ! -d $TMP_DIR/${system}/q${i} ; then
>          mkdir $TMP_DIR/${system}/q${i}
>          cp -r $TMP_DIR/${system}/${system}.* $TMP_DIR/${system}/q${i}
> fi
> 

Are the files really there? If all the files are in the directory 
$TMP_DIR/${system}/q${i} there is no reason why the ph.x code stops.
If they are not, I cannot help you. It is not a problem of ph.x.


> BTW, could you please describe more in detail about the newly added 
> information
> in INPUT_PH.html for distributing phonon calculations to cluster? If I set 
> wf_collect
> to true, there should be no relation in nproc and npool between pw.x run and 
> later
> two ph.x runs, right? Taking AlAs in the GRID_example as example, If I want 
> to use
> server with 8 CPU core to do calculations for each one q point (8 servers in 
> total),
> what are the values of images and pools?
> 
> 
You are right, the explanation refer only to the case
wf_collect=.false.. However, image parallelization of ph.x is very
experimental, so be patient. At the moment it divides both q and irreps.
Load balancing on q only is not implemented.  


Andrea 



> Huiqun Zhou
> @Earth Sciences, Nanjing University, China
> 
> ----- Original Message ----- 
> From: "Dal Corso Andrea" <dalcorso at sissa.it>
> To: "PWSCF Forum" <pw_forum at pwscf.org>
> Sent: Tuesday, July 06, 2010 4:14 PM
> Subject: Re: [Pw_forum] Distributing phonon calculations todifferent 
> machines
> 
> 
> > On Tue, 2010-07-06 at 12:13 +0800, Huiqun Zhou wrote:
> >> Sorry, I sent an unfinished message.
> >>
> >> When using _ph0{prefix}.phsave, I got the error message shown in the
> >> previous
> >> message.
> >>
> >> Here is the snippet of my script for distributing lsf tasks:
> >>
> >> ......
> >> nq=`sed -n '2p' ./${system}_q${nq1}${nq2}${nq3}.dyn0`
> >>
> >> for ((i=1; i<=$nq; i++))
> >> do
> >>     if test ! -d $TMP_DIR/${system}/q${i} ; then
> >>         mkdir $TMP_DIR/${system}/q${i}
> >>         cp -r $TMP_DIR/${system}/${system}.* $TMP_DIR/${system}/q${i}
> >>     fi
> >>     if test ! -d $TMP_DIR/${system}/q${i}/_ph0${system}.phsave ; then
> >>         mkdir $TMP_DIR/${system}/q${i}/_ph0${system}.phsave
> >>         cp -r $TMP_DIR/${system}/_ph0${system}.phsave/*
> >> $TMP_DIR/${system}/q${i}/_ph0${system}.phsave
> >>     fi
> >> done
> >>
> >> for ((i=1; i<=$nq; i++))
> >> do
> >> cat > ${system}_q${i}.in << EOF
> >> phonons of ${system}
> >>  &inputph
> >>   tr2_ph = 1.0d-13,
> >>   alpha_mix(1) = 0.2,
> >>   prefix = '${system}',
> >>   ldisp = .true.,
> >>   recover = .true.
> >>   nq1 = ${nq1}, nq2 = ${nq2}, nq3 = ${nq3}
> >>   start_q = $i, last_q = $i
> >>   outdir = '$TMP_DIR/${system}/q${i}',
> >>   fildyn = '${system}_q${nq1}${nq2}${nq3}.dyn'
> >> ......
> >> EOF
> >> $ECHO "calculation of q point $i"
> >> bsub -a intelmpi -n $processes \
> >>      -R "span[ptile=8]" \
> >>      -J ${r}q${i}anda \
> >>      -oo ${system}_q${i}.out \
> >>      -eo ${system}_q${i}.err \
> >>      $PH_COMMAND -input ./${system}_q${i}.in
> >> done
> >>
> >>
> >> Huiqun Zhou
> >> @Earth Sciences, Nanjing University, China
> >>         ----- Original Message ----- 
> >>         From: Huiqun Zhou
> >>         To: pw_forum at pwscf.org
> >>         Sent: Tuesday, July 06, 2010 12:00 PM
> >>         Subject: [Pw_forum] Distributing phonon calculations to
> >>         different machines
> >>
> >>
> >>         dear developers,
> >>
> >>         Please clarify what directory should be copied
> >>         for distributing phonon calculations
> >>         to different machines,  _ph{prefix}.phsave
> >>         or _ph0{prefix}.phsave? The former is
> >>         described in the manual INPUT_PH.html, the latter is used in
> >>         the GRID_example.
> >>         Although there is no _ph{prefix}.phsave existed after the
> >>         preparatory run with
> >>         start_irr=0 and last_irr=0, using the former works OK at the
> >>         cost of redundant
> >>         calculations.
> >>
> >>              Representation #  1 mode #   1
> >>
> >>              Self-consistent Calculation
> >>
> >>          %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
> >>         %%%%%%%%%%%%%%%%%
> >>              from davcio : error #        25
> >>              error while reading from file
> >>          %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
> >>         %%%%%%%%%%%%%%%%%
> >>
> >>              stopping ...
> >
> > Thank you for the message. I will correct the INPUT_PH documentation.
> > The correct directory is _ph0{prefix}.phsave.
> >
> > This message usually means that you have not copied all the required
> > files. Did you copy all the files produced by pw.x?
> >
> > HTH,
> >
> > Andrea
> >
> >
> >
> >>
> >>         ______________________________________________________________
> >>
> >>         _______________________________________________
> >>         Pw_forum mailing list
> >>         Pw_forum at pwscf.org
> >>         http://www.democritos.it/mailman/listinfo/pw_forum
> >> _______________________________________________
> >> Pw_forum mailing list
> >> Pw_forum at pwscf.org
> >> http://www.democritos.it/mailman/listinfo/pw_forum
> > -- 
> > Andrea Dal Corso                    Tel. 0039-040-3787428
> > SISSA, Via Beirut 2/4               Fax. 0039-040-3787528
> > 34151 Trieste (Italy)               e-mail: dalcorso at sissa.it
> >
> >
> > _______________________________________________
> > Pw_forum mailing list
> > Pw_forum at pwscf.org
> > http://www.democritos.it/mailman/listinfo/pw_forum
> > 
> 
> 
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
-- 
Andrea Dal Corso                    Tel. 0039-040-3787428
SISSA, Via Beirut 2/4               Fax. 0039-040-3787528
34151 Trieste (Italy)               e-mail: dalcorso at sissa.it





More information about the users mailing list