2008/6/6 Ilja Livenson <[log in to unmask]>:
> Hello,
>
> we have 2 sets of nodes (amd and intel based) that run torque and have the
> same torque head node that is also CE. The distinction is done on the queue
> level - some queues use only nodes from amd set, other queues only from
> intel set. Now the question is what is the correct way to publish this sort
> of information to bdii?
Hi Ilja,
The short answer for your case is don't bother to publish the
difference and just publish the dominent
CPU set since they are both binary compatible. There are many sites
publishing this
kind of mixture of nodes as a single site.
The other options are to publish 1 (in fact2) GlueClusters and 2
GlueSubclusters.
You can do this with
1) Variation on the outdated instructions here though the principal
still works and other sites are publishing
low/high memory subclusters or ia64/x86_64 subclusters.
http://goc.grid.sinica.edu.tw/gocwiki/How_to_publish_different_memory_limits_for_different_queues_on_the_same_CE
2) Run two CEs.
3) Wait for the work of the WNWorkingGroup for which exactly this kind
of thing is it mains concern.
https://twiki.cern.ch/twiki/bin/view/EGEE/WNWorkingGroup
but do not wait for this last one if you have hardware sitting in your
machine room now. Things are progressing
here with bits of prototype yaim now in place but there are still
gaps in things that are needed.
Steve
> Also, there's a question about accounting: we are using dgas for that and
> it's not obvious how to configure it to handle different queues differently.
>
> Anyway, I was wondering if anyone has had the same problem or might provide
> some advice.
>
> thanks,
> Ilja
>
--
Steve Traylen
|