Date: Thu Dec 18 2014 - 01:43:11 CST
Interesting. The last paragraph and the attached file seem to be missing from my previous mail. I just paste these here.
I have paste a script below, I have used to start a series of MD simulations. I have also included a namd config file to show how I passed environment variables down to the jobs. Something that I always found to be very cumbersome on the BlueGenes.
# submit jobs on BlueGene via LoadLeveler
tasks=`expr $nodes \* $threads`
if [ -z "$1" -o -z "$2" ]; then
echo "Usage: $0 start stop"
for step in `seq $start $stop`; do
curr=md`printf "%03i" $step`
prev=`expr $step - 1`
prev=md`printf "%03i" $prev`
if [ $step -ne $start ]; then
dependency="#@ dependency = ($prev == 0)"
cat << _EOF
#@ job_name = $curr
#@ step_name = $curr
#@ job_type = bluegene
#@ bg_size = $nodes
#@ node_usage = shared
#@ wall_clock_limit = 04:00:00,03:58:00
#@ output = $mycwd/$curr.out
#@ error = $mycwd/$curr.err
#@ notification = error
#@ arguments = -np $tasks -mode VN -cwd $mycwd -env "NAMD_PREV=$prev NAMD_CU
RR=$curr" -exe $binary -args $input
======= config ======
# set file names
set previous $env(NAMD_PREV)
set current $env(NAMD_CURR)
From: owner-namd-l_at_ks.uiuc.edu [owner-namd-l_at_ks.uiuc.edu] on behalf of hannes.loeffler_at_stfc.ac.uk [hannes.loeffler_at_stfc.ac.uk]
Sent: 18 December 2014 07:36
Subject: RE: namd-l: setting up namd simulation in IBM Blue Gene/L
We have retired our BlueGenes L and P a while ago so I can't help you much.
What I remember is that our LoadLeveler on the L never had scripting capabilities so your final mpirun would have simply be ignored. The user was supposed to use #@arguments and possibly #@executable. There was a default for the latter. In any case, make sure that you are calling the right executable here if you have to do it explicitly (may not even be mpirun on the L)
This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:21:28 CST