[torquedev] "any" resources
mgutteri at fhcrc.org
Fri Dec 16 10:45:41 MST 2005
Caveat- we don't run a mixed environment- yet.....
Anyway, I would suggest something different. When one requests nodes
via the "nodes=" resource, torque seems to operate on the principal that
it should choose from the widest possible set of nodes. That is, if I
only say "nodes=1", it'll choose from the largest set, versus
"nodes=1:foo", which obviously chooses from a smaller set.
It would make sense to me (it was my first reaction) that if I don't
select an architecture, it should choose from any available arch.
However, pbs_resources(7B) indicates that arch defaults to "PBS_MACH",
which, I guess, is why you've set defaults on those queues.
I'm perhaps overreaching, but it seems to me that if what you're wanting
is a desirable behaviour, it would be better to change the defaulting
mechanism for the "arch" resource, vis if arch isn't set, it can run on
Otherwise my vote is for "any". Splat (*) is too hard to use in scripts
On Thu, 2005-12-15 at 21:49 -0800, Garrick Staples wrote:
> We've always seperated our 32bit and 64bit Linux nodes into different
> queues. Now I'd like use queues with mixed nodes. I have every node
> with the "arch" resource as either i686 or x86_64 and are starting to
> set a default arch on queues, letting the user override the default.
> So something like:
> set queue main resource_default.arch=i686
> set queue foo resource_default.arch=x86_64
> Now we need to let users specify that their job can run on either arch.
> I guess what we need is to establish a standard name for an "any"
> Any thoughts on good solutions?
> torquedev mailing list
> torquedev at supercluster.org
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://www.supercluster.org/pipermail/torquedev/attachments/20051216/23cefbd8/attachment.bin
More information about the torquedev