Author Topic: Using NumForce with parallel egrad in TM 6.3  (Read 4213 times)

martijn

  • Full Member
  • ***
  • Posts: 53
  • Karma: +0/-0
Using NumForce with parallel egrad in TM 6.3
« on: June 07, 2011, 12:41:10 PM »
Hi,

I'm interested to use NumForce in combination with the new parallel version of egrad in TM 6.3 so I can calculate the frequencies of an an excited state structure of a large molecule/cluster. When I however attempt to do this after setting the following environment settings in my run script:

#set up TM environment
export PARA_ARCH=SMP
export TURBODIR=/work/e05/e05/martijn/TURBOMOLE
export PARNODES=24
source $TURBODIR/Config_turbo_env

(which are btw the same I use for Jobex) NumForce dies before doing anything with the following error message:

 running on nid01534
 date: Sun May 15 23:38:19 BST 2011

error reading machinefile /var/spool/PBS/aux/218559.sdb
Application 607424 exit codes: 1
Application 607424 resources: utime ~0s, stime ~0s

It seems like TM is looking for the machinefile one would give if one would use NumForce parallel in "task farming" mode and is thus confused about what kind of parallelisation I attempt to do. Can NumForce in TM 6.3 not exploit the shared memory parallisation in dscf and egrad?

I'm aware of the task farming method but to efficiently run the frequency calculations on my molecules I would like to in the end exploit a combination of both (where NumForce distributes the egrad calculations to different nodes, each with 24 cores, where the egrad calculations run in shared memory mode exploiting all the cores on the node).

Thanks in advance,

Martijn