vasp stalls across nodes

Message

internetscooter · #1 Post by **internetscooter** » Tue Jan 10, 2012 11:43 pm

Hi All,

I am not a molecular dynamics expert but I am assisting Monash VASP users with a problem they have running on our HPC system.

VASP 5.2.11 will run fine on one node (12 cores) but when split across more than one node it often stalls with all processes running at 100% but no output. strace shows that it is on a polling loop waiting for "something".

Are there any verbose output options etc to allow me to work out what is going wrong? Is there any special build configuration required for it to work across multiple nodes?

Any tips would be much appreciated.

Cheers,

Paul

internetscooter · #2 Post by **internetscooter** » Fri Jan 13, 2012 9:35 am

should 5.2.11 be run with "-tasks N" instead of mpirun?

Referring to this info, which suggests the way VASP is started has changed
http://www.lrz.de/services/software/chemie/vasp/

We are still seeing completely random behaviour, with about a 20/80 chance of a run actually working.

Cheers,

Paul

alex · #3 Post by **alex** » Fri Jan 13, 2012 10:02 am

Hi Paul,

it would be helpful if you details like:

- which platform you are on
- your type of network (Infiniband etc.)
- mpi-version
- compiler
- ...

Thanks

alex