LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Alessandro Paolini
said:
> some update ago has been created the yaim variable
> CLUSTER_HOST to allow
> the creation of more sublusters under the same CE, but it is
> still not
> usable (it must be set to CE_HOST for the time being).
> We know that WMS isn't able to handle more suclusters yet, so
> we'd like
> to know if there is the idea to use this new variable in a
> short term or
> in a long term; we've also known from WMS developers that no request
> (till now) has been done to them to modify properly the WMS
> to make the
> service handle more sublusters under a CE (a changement not
> so complex)
There are two different questions here, depending on whether by "CE" you
mean a glue CE (i.e. a batch queue) or a front-end host (gatekeeper).
What the WN working group was trying to do was to get YAIM to configure
multiple subclusters (and clusters) on the same host, but connected to
different glue CEs. That is supported by the WMS and in fact many sites
already do it, e.g. to have large-memory and small-memory queues, but
they have to configure it by hand because YAIM doesn't support it.
There is a separate question of having multiple subclusters for the
same glue CE, i.e. you would submit to a single queue and the WMS would
tell the batch system which subcluster(s) it matched against. That has
been something people have asked for for many years, but indeed I don't
think it has ever been formally on the work plan. In the present
situation I wouldn't expect any change before the end of EGEE, if it
happens at all it will be in the context of EGI/EMI and probably using
the glue 2 schema, since the WMS will have to change anyway to use that.
Stephen
--
Scanned by iCritical.
|