[cig-commits] r13920 - cs/portal/trunk/northridge/backend

[leif at geodynamics.org]

Author: leif
Date: 2009-01-22 12:57:34 -0800 (Thu, 22 Jan 2009)
New Revision: 13920

Modified:
   cs/portal/trunk/northridge/backend/mineos.py
Log:
Specify queue and CPU count when scheduling "waiter" job (which
doesn't seem to be working anyway).

The glacial pace of the spheroidal computation turns out to be a
problem with 64-bit Mineos.  On both Lonestar and CITerra, it reaches
a certain point and then appears to hang, until it is ultimately
killed by LSF.  Compiling with "-m32" makes the problem go away.  On
Lonestar, with this work-around in place, spheroidal takes just over
three hours.


Modified: cs/portal/trunk/northridge/backend/mineos.py
===================================================================
--- cs/portal/trunk/northridge/backend/mineos.py	2009-01-22 16:07:42 UTC (rev 13919)
+++ cs/portal/trunk/northridge/backend/mineos.py	2009-01-22 20:57:34 UTC (rev 13920)
@@ -239,7 +239,7 @@
         # This assumes the job name is unique.  We could slurp the job
         # ID from the 'bsub' output... but could that really be
         # considered more robust?
-        argv = ['bsub', '-J', 'waiter', '-K', '-W', '0:01',
+        argv = ['bsub', '-q', 'serial', '-n', '1', '-J', 'waiter', '-K', '-W', '0:01',
                 '-w', 'done(%s)' % jobArrayName,
                 '/bin/true']
         status = os.spawnvp(os.P_WAIT, argv[0], argv)