problem running parallel version of vasp

Questions regarding the compilation of VASP on various platforms: hardware, compilers and libraries, etc.

Moderators: Global Moderator, Moderator

Post Reply
Message
Author
slaubach
Newbie
Newbie
Posts: 5
Joined: Tue Nov 23, 2004 12:16 pm

problem running parallel version of vasp

#1 Post by slaubach » Thu Nov 23, 2006 12:59 pm

I try to run the parallel version of vasp4.6 using mpich integrated in the sun grid engine.
Compiling the prallal version gave no errors.
I set NPAR=4 and tried to run on 4 Nodes.

When trying to run I get an Error
Error reading item 'IMAGES' from file INCAR
I thought the IMAGES flag was only for elastic band calculation.

I don´t know if I didn´t compile the file right(wrong editing of makefile) or if there is a INPUT problem.

Thank you in advance for any help
Last edited by slaubach on Thu Nov 23, 2006 12:59 pm, edited 1 time in total.

slaubach
Newbie
Newbie
Posts: 5
Joined: Tue Nov 23, 2004 12:16 pm

problem running parallel version of vasp

#2 Post by slaubach » Thu Nov 23, 2006 1:39 pm

I also tried with NPAR=1 on one node and got the same error
Last edited by slaubach on Thu Nov 23, 2006 1:39 pm, edited 1 time in total.

admin
Administrator
Administrator
Posts: 2922
Joined: Tue Aug 03, 2004 8:18 am
License Nr.: 458

problem running parallel version of vasp

#3 Post by admin » Thu Nov 23, 2006 3:05 pm

you are right, IMAGES is EXPLICITELY read only for NEB runs. However, reading IMAGES is the very first read attempt from any file in the parallel version of vasp. So mostly this error indicates that the INCAR file is missing (eg. in the wrong directory) or corrupted. If IMAGES is not set at all, there is no error statement. Please check if the INCAR file is correctly transferred to the master node of your parallel job.
Last edited by admin on Thu Nov 23, 2006 3:05 pm, edited 1 time in total.

taylor
Newbie
Newbie
Posts: 1
Joined: Thu Nov 30, 2006 1:05 pm
License Nr.: support for groups at UFL

problem running parallel version of vasp

#4 Post by taylor » Thu Nov 30, 2006 1:13 pm

I've experienced this problem as well and found, in my case, that it was caused by the presence of the -i8 option in the vasp.4.6.lib makefile where the rdatab subroutine is built.

My VASP is built for the Opteron (64 bit) platform using mvapich and the Pathscale Compiler.
Last edited by taylor on Thu Nov 30, 2006 1:13 pm, edited 1 time in total.

alemottura
Newbie
Newbie
Posts: 1
Joined: Fri Mar 23, 2007 1:44 pm
License Nr.: jk35QGB7
Location: London

problem running parallel version of vasp

#5 Post by alemottura » Tue Mar 27, 2007 1:23 pm

Hi,

I am facing exactly the same problem when running in parallel (not in serial). I have narrowed it down to the fact that VASP looks for the INCAR file locally on every secondary node assigned to my jobs. PBS (as far as i know) only copies the files from my home to the master node for my calculations (and I am 100% positive this happens correctly)...therefore I find myself having to use some fiddly utilities to copy the files to every secondary node assigned to my job.

I find this is not the case for other parallel software...and I know of other people using VASP on different machines without this problem...so I guess it's something about the way the parallel version is compiled.

Would taking away this -i8 flag from the vasp.4.6.lib makefile solve this problem? or am I facing something different?

Sorry for asking again about this...but I wanted to know something more before I ask to recompile VASP.
Last edited by alemottura on Tue Mar 27, 2007 1:23 pm, edited 1 time in total.

admin
Administrator
Administrator
Posts: 2922
Joined: Tue Aug 03, 2004 8:18 am
License Nr.: 458

problem running parallel version of vasp

#6 Post by admin » Wed Mar 28, 2007 10:16 am

whether of not the input files have to be copied to each of the parallel nodes depends on the setup of your parallel architecture. This is not vasp related. Please contact your local system administrator concerning this
It has nothing to do with the -i8 flag (which gives the bit-length of integer numbers).
Please note however that eg mpich uses the standard C-declaration of integers (32bit). Hence if you use mpich as paralellization parser, you must not set -i8 explicitely, because the MPI variables will not be communicated properly between the nodes then
Last edited by admin on Wed Mar 28, 2007 10:16 am, edited 1 time in total.

Post Reply