On 20/06/11 14:44, Daniela Bauer wrote:
>> change some details so "GlueCEUniqueID: ce01...." becomes
>> "GlueCEUniqueID: cream02...".
> How would yaim have ever heard of ce01 on cream02 ? It should make the
> ldif files automatically.
I had a look again at this and the overall issue is that in
'site-info.def' one cannot put a list as the value of 'CE_HOST',
even if some recent pages talk about a new node type called
'glite-CLUSTER':
http://glite.cern.ch/glite-CLUSTER/
http://www.gridpp.ac.uk/deployment/users/Cluster-node.pdf
The old-style technique that I have seen documented in some web
page is to manually modify the 'CE_HOST' variable before running
YAIM on a particular CE.
But this page at DESY documents an alternative:
http://grid.desy.de/install/glite.html
> /opt/glite/yaim/bin/yaim -v -s /opt/misc/yaim/`hostname
-s`-info.def -n lcg-CE -n TORQUE_utils
Inspired by the latter I have decided to do something similar:
to have 'site-info-ce01.def', ... 'site-info-cream02.def' where
each contains something like:
source "$GLITE_LOCATION/yaim/etc/site-info.def"
CE_HOST=cream02.$MY_DOMAIN
I had also created a 'services/batch-torque' file that is
'source'd by 'site-info.def' but now I am thinking of
'source'ing it directly from the CE-specific 'site-info-*.def'
or some similar variant.
> As fo the bdii, all you do is give it a list of CEs (I assume you
> already do that for the lcgCEs, there's no difference with cream).
I guess that here you refer to 'BDII_REGIONS'.
> [ ... ] each CE only publishes a share of the cluster
> (i.e. I have 400 CPUs and each of my 4 CEs publishes 100), but of
> course I still get each running job reported 4 times, but so does
> everybody else.
That sounds reasonable, as well as StephenJ's suggestion about
publishing averages. I have also received another suggestion to
publish the cluster via a single CE.
Many thanks to all those who have made suggestions so far.
|