Page 1 of 1

VTST and no output?

Posted: Sun Aug 03, 2008 6:13 pm
by mehmetb
Hello,

I have finally installed TST version of VASP on our MacOSX cluster. However, when I try to run it, it just produces a blank output and terminates. No errors, no warnings:

Code: Select all

n99(0):  POSCAR, INCAR and KPOINTS ok, starting setup
n99(0):  WARNING: wrap around errors must be expected
n99(0):  FFT: planning ... 1
n99(0):  reading WAVECAR
All nodes terminated successfully.
Some important points:
* The VASP without TST was working correctly
* The gamma point version (-DwNGZhalf ) worked OK (but that's not we want)
* When I use -DNGZhalf for the parallel multi k-point version, it fails.
* I compiled all TST-related files with NO optimization. The result was the same.

I would really appreciate your suggestions. We are looking forward to see TST running on our cluster!!

-Memo

Re: VTST and no output?

Posted: Sun Aug 03, 2008 10:11 pm
by graeme
This is very strange.

Are you really saying that your vasp installation without the TST code works without any problems for the same calculation that crashes with the TST code? The multi k-point parallel run, for example, runs fine? Is the makefile the same except for adding the TST objects?

Are you running a regular (minimization or dynamics) calculation when the code quits? If so, this is very strange, because the VTST code should only be adding a few print statements to the output if you are not doing a transition state calculation or using one of our optimizers.

If I were guessing, I would suggest that it's hitting a memory limitation. If these are older PPC machines, the xlf compiler has a flag to unset the stack and data limits. It can also be done with the limit shell command on both PPC and Intel machines:

limit stacksize unlimited
limit datasize unlimited
limit filesize unlimited

It is crashing near the point when the major memory is allocated. But this is easy to test with a small calculation, and it also doesn't make sense that the VTST code is affecting it -- almost no additional memory is allocated in our modules.

One last suggestion: make sure that the chain.o and chain.mod files are removed before building vasp. We do replace the chain.F file, and if you have old chain objects, there could be problems. Doing a clean build would test this possibility.

If you still have problems, we could test your makefile and run on a comparable OSX machine here.

Good luck!