[gold-users] Connecting up Gold with Moab on a Cray...

Michael Sternberg sternberg at anl.gov
Fri Jun 4 12:02:45 MDT 2010


On Jun 4, 2010, at 12:40 , ingmar wrote:
> Our supercomputing group and I are going to get Gold (running on a 
> separate Linux box) connected to Moab running on a Cray.  This is 
> probably a rather rare configuration.  I was wondering if anyone out in 
> the community has done such a thing and what problems were encountered.
> Thanks,
> -Ingmar Thompson
> NCAR

Ingmar,

While not using a Cray, I do run gold and moab on different servers (having to do with failover).  The Moab integration is very straightforward, e.g.:

AMCFG[bank]     SERVER=gold://Linuxbox/ TIMEOUT=15
AMCFG[bank]     CHARGEPOLICY=DEBITALLBLOCKED
AMCFG[bank]     NODECHARGEPOLICY=MAX
AMCFG[bank]     JOBFAILUREACTION=HOLD
AMCFG[bank]     FLUSHINTERVAL=12:00:00


Normally gold gets sent the number of cores and CPU time and calculates charges based on its "machine"-related attributes.  For finer control (I have different node generations) I use this:


# http://www.clusterresources.com/products/mwm/docs/6.4allocationmanagement.shtml#flags
# To use node-specific CHARGERATE as defined below, set LOCALCOST
AMCFG[bank]     FLAGS=LOCALCOST
...
NODECFG[n143]   RACK=5  SLOT=36 PROCSPEED=2667  SPEED=1.0 CHARGERATE=0.5
NODECFG[n144]   RACK=6  SLOT=36 PROCSPEED=2667  SPEED=1.0 CHARGERATE=0.5

NODECFG[n145]   RACK=9  SLOT=1  PROCSPEED=2533  SPEED=2.5 CHARGERATE=1.0
NODECFG[n146]   RACK=10 SLOT=1  PROCSPEED=2533  SPEED=2.5 CHARGERATE=1.0
...


With best regards,
Michael



More information about the gold-users mailing list