note on scratch (Re: [Wien] troubles in LAPW1)

Torsten Andersen Torsten.Andersen at Fysik.UU.se
Fri Jul 11 09:36:22 CEST 2003


Yeah, in principle you are right. However, the example you give is a 
waste of cpu time... as you can do convergence for RKM7 first, then 
simply change to RKM8 in case.in1c and restart. Typically, the RKM8 case 
will then converge in 3 cycles if your potential is converged.

However, something along the line of

---
#!/bin/csh -f
set workdir = /path/to/case
set SCRATCH = /tmp/scratch_1
set remote = rsh
foreach machine (`awk -F: '{print $2}' $workdir/.machines`)
 ($remote $execmachine "mkdir $SCRATCH; other_commands")
end
cd $workdir
runsp_lapw -ec 0.0001 -so -p
foreach machine (`awk -F: '{print $2}' $workdir/.machines`)
 ($remote $execmachine "rm -rf $SCRATCH")
end
----

in a c-shell script will likely solve that problem. Then use the shell 
script as your base for calculations. If you are using a queue system, 
you are likely to have to make a kind of shell script anyway. Be careful 
with the last "foreach" - it will remove all your scratch directories 
and their content, but I guess that was what you wanted.

Have fun,
Torsten Andersen.

Steven Homolya wrote:

>An note on SCRATCH choice, perhaps of importance for some:
>
>If $SCRATCH is a global directory like say /tmp, you will not be able to run 
>two calculations that have the same case name simultaneously. (Took me quite 
>a while to figure out why wien2k was crashing a lot.) Setting SCRATCH to a 
>subdirectory of the wien2k case directory, e.g. SCRATCH="./wien2k_tmp", will 
>fix this problem. (But then one must remember to create the scratch directory 
>for every new case... and cleaning up gets more cumbersome.)
>
>Eg. Say you want to run two instances of the TiC example one with RKmax = 7.0 
>and another with RKmax = 8.0, and you set them up in
>
>~/RKM7/TiC
>~/RKM8/TiC
>
>and start the scf cycles for both. If SCRATCH is an absolute path like /tmp, 
>the two calculations will confuse each others O/Ps and I/Ps, and crash or 
>produce nonsense. No problems if each case has its own scratch dir.
>
>  
>

-- 
Dr. Torsten Andersen
Department of Physics, Condensed Matter Theory Group, Uppsala University
UU-WWW: http://www.fysik4.fysik.uu.se/    TA-WWW: http://deep.at/myspace






More information about the Wien mailing list