LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Jeff Templon
said:
> Consider a more realistic case, where you are requesting 240
> normalized
> CPU minutes. One of the many 72-hour 'long' queues will
> satisfy such a
> request. What happens now is that the job is submitted to the LRMS
> without informing the LRMS that the job will only take 240 normalized
> minutes.
Not only that, depending on the ranking the job may go to the 72-hour
queue even though the site has a 6-hour queue which would have been more
suitable (although that may be fixed by a better ERT ...)
One issue is the way the hardware is represented in the Glue schema.
The way it's structured now is that you have subclusters which define
the characteristics like cpu speed, memory size etc, where the WNs
behind each subcluster should all be identical. If the job matches its
requirements against a particular subcluster you would need to tell the
LRMS which subcluster(s) matched. Each subcluster has a name, so
potentially you could just pass through subcluster names as long as they
could be matched against some kind of property name in the LRMS.
Stephen
|