On Tue, 18 Jan 2005, Burke, S (Stephen) wrote:
> LHC Computer Grid - Rollout
>> [mailto:[log in to unmask]] On Behalf Of Mattias Wadenstein
> said:
>> Well, consider it a temporary hack until the software has
>> better support
>> for different pbs site setups. For now, the assumption that
>> each vo has
>> its own queue seems to be all over the code, and the pbs
>> interface was a
>> much smaller "area" to cover than the interface to the
>> information system.
>
> As we've recently been discussing, this is supposed to be a production
> system, not a testbed, so I don't think hacks are appropriate, even
> temporarily. Equally, the configuration tools obviously should not assume
> any particular queue setup (the queue-per-VO is itself largely a hack and
> not in my view a particularly useful one).
All software installs require tweaking and debugging to make it work in
the local situation. This should be the object of the functional testing
before being accepted in into production use (and once you are there, no
new hacks of course).
The alternative in this instance would be to tweak pbs into giving a false
apperance of multiple queues, despite all jobs being scheduled in one
single queue. Or to fix it properly, to take out the assumption of
queue-per-VO, but that is a large and complex operation that I would
rather leave to the middleware developers.
/Mattias Wadenstein
|