Posted by Huub van Dam on December 02, 2004 at 09:30:46:
Hi,
I am interested in capability class CASPT2 calculations of transition metal complexes. To get a feel for what can be done I have build MOLCAS 6.0 on HPCx (see below for more details on the machine).
When I run a test calculation on 32 processors everything is fine. However when I run the same calculation on 64 processors SEWARD exits with the following error message:
*******************************************************************************
*******************************************************************************
*** ***
*** ***
*** Location: AixWr ***
*** File: ORDINT ***
*** ***
*** ***
*** Premature abort while writing buffer to disk: ***
*** Condition: rc != LenBuf ***
*** Actual : 0!= 262144 ***
*** ***
*** ***
*******************************************************************************
*******************************************************************************
0:0:Abend terminating!:: 97
0:ARMCI aborting 97(61)
The machine I am using is a cluster of 32-way IBM power4 shared memory nodes. The machine has a parallel filesystem such that all nodes see the same directory structure both in the home directories and the scratch directories. The configure options I used are '-parallel lapi -par_type distributed'. Can anyone shed some light on why SEWARD runs fine on a single node but fails on two nodes? Have I chosen the right approach to build MOLCAS on the type of platform I described?
Thanks in advance,
Huub