Testbed Support for GridPP member institutes
> [mailto:[log in to unmask]] On Behalf Of John Walsh said:
> I believe that each batch queue could be declared a
> a subcluster, so
> that the correct information is also published
> (including OS etc).
> [Stephen (Burke), is my assumption correct?]. This
> may require a change
> to the GIP configuration. I have not implemented a
> subcluster solution
> as the nodes are a similar SPEC to those published by
> the site.
The mapping is supposed to be (1 or more subclusters) -> (1 cluster) ->
(1 or more CEs). Unfortunately the broker can't cope with more than 1
subcluster per cluster, so if you want to publish like that you have to
duplicate the cluster as well and have a separate set of queues. You can
however keep everything attached to a single CE head node. This has been
done by Steve Traylen:
http://goc.grid.sinica.edu.tw/gocwiki/How_to_publish_different_memory_li
mits_for_different_queues_on_the_same_CE
The main downside is that if you wanted to represent lots of different
subcluster configurations you would end up with a huge explosion of
queues (atlas_short_bigmem_fast-cpu_debian ...) There is supposed to be
a working group looking at doing this more efficiently, but don't hold
your breath :)
> It might be possible to publish that this subcluster has no
> outbound IP access (again Stephen, can you verify this?).
You can publish it, but there's no guarantee that people will check it
in their JDL as it's generally assumed to be there. Also various bits of
the middleware assume outbound connectivity.
Stephen
|