junk files: what are they
Gets Around
2:46:03 AM PST - Tue, Nov 15th 2011
Hi,

i'm running nwchem installed from RPM on CentOS 5 x86_64:
(http://nwchemgit.github.io/Special_AWCforum/st/id262/RPMS_of_NWchem.html), with
mpiexec -np 4 nwchem
I want to limit disk io as much as possible (see the attached input below), but still
noticed periodically the presence of "*.junk.[0-3]" files.
For the CH3Cl system one junk file has a size of ~10 Megabytes.
What are the junk files are how to get rid of them?

geometry noautoz noautosym
C 8.891415 8.514659 8.352891
Cl 8.891415 8.514659 10.130231
H 8.891415 9.543977 8.0
H 9.78283 8.0 8.0
H 8.0 8.0 8.0
end
basis spherical
  • library aug-pc-4
end

dft
mult 1
xc slater perdew81
iterations 300
convergence gradient 0.0005
convergence energy 1e-06
convergence density 1e-05
grid fine nodisk
smear 0.0
tolerances tight
direct
noio
end

set int:txs:limxmem 134217728

memory total 8000 Mb

task dft energy gradient

Forum Vet
2:47:36 PM PST - Fri, Nov 18th 2011
These files are generated at startup to test disk availability and should disappear (they are deleted by the code) once the DFT starts running. Disk is otherwise not used in your calculation.

Bert




Quote:Marcindulak Nov 15th 9:46 am
Hi,

i'm running nwchem installed from RPM on CentOS 5 x86_64:
(http://nwchemgit.github.io/Special_AWCforum/st/id262/RPMS_of_NWchem.html), with
mpiexec -np 4 nwchem
I want to limit disk io as much as possible (see the attached input below), but still
noticed periodically the presence of "*.junk.[0-3]" files.
For the CH3Cl system one junk file has a size of ~10 Megabytes.
What are the junk files are how to get rid of them?

geometry noautoz noautosym
C 8.891415 8.514659 8.352891
Cl 8.891415 8.514659 10.130231
H 8.891415 9.543977 8.0
H 9.78283 8.0 8.0
H 8.0 8.0 8.0
end
basis spherical
  • library aug-pc-4
end

dft
mult 1
xc slater perdew81
iterations 300
convergence gradient 0.0005
convergence energy 1e-06
convergence density 1e-05
grid fine nodisk
smear 0.0
tolerances tight
direct
noio
end

set int:txs:limxmem 134217728

memory total 8000 Mb

task dft energy gradient

Gets Around
4:43:14 AM PST - Sat, Nov 19th 2011
It seems that the junk files are generated at each SCF interation step and I guess this results in a zero CPU load during such IO. For small molecules nwchem may spend substational amount of time writing those junk files, but i haven't really analysed that. What I see is CPU load zero with the `top` command almost all the time in such cases. Could you confirm that?

Gets Around
8:59:12 AM PST - Mon, Dec 5th 2011
I trying to illustrate the problem with junk files.
Please use my nwchem input above. You can use a smaller basis set and grid.
Before you start the calculation, start the following script in the same directory:

#!/bin/sh
file="*.junk.0"
laststat='0'
while /bin/true;
do
t=`date +%s.%N`
if test -e ${file} 2> /dev/null;
then
newstat='1'
else
newstat='0'
fi
if [ "${newstat}" != "${laststat}" ];
then
echo ${t} ${newstat} >> junk_count
laststat=${newstat}
fi
done

This script will put a line into the junk_count file every time a *.junk.0 is created or deleted,
and supposedly will show that junk files are created/deleted more than once.
I used http://nwchemgit.github.io/images/Nwchem-6.0-binary-redhat-5-5-gcc-4-1-2.tar.bz2 on a x86_64 CentOS 5 system.
If there is no way to stop those junk files from being created please consider as a priority fixing that.
IO is IO no matter how you call it.


Forum >> NWChem's corner >> Running NWChem