On May 28, 2008, at 2:20 AM, Jan Ploski wrote:
I would say that treating Globus job submissions like an extended
batch
queue should be among the allowable use cases. Users of a local batch
scheduler may view the "Grid" as a drop-in replacement, which they
expect
to be at least as easy to use and efficient, just more fault-
tolerant and
scalable.
Interesting point of view. I agree that this is how many users
(especially first-time ones) expect it to work, irrespective of how
it is actually wired and plumbed. Thanks very much for the input.
Learning to run at scale is clearly one of the interesting challenges
for grid software of any kind in the near-term future. For the
moment, I think the user-throttling advice is the right starting
point -- don't start zillions of jobs and just expect them to be
totally idle in the queue until they start running on the remote
resource (as you also agreed). Whether we can or should design
things so that this use case can be accommodated in the future is an
interesting challenge. I would say that most designers of software
from the users' point of view would agree with you that the answer
should be "yes" where it is definitely "no" at the present time --
instead, one of the user solutions I pointed to should be used to
prevent lots of outstanding pending processes.
Alan
Alan Sill, Ph.D
TIGRE Senior Scientist, High Performance Computing Center
Adjunct Professor of Physics
TTU
====================================================================
: Alan Sill, Texas Tech University Office: Admin 233, MS 4-1167 :
: e-mail: [EMAIL PROTECTED] ph. 806-742-4350 fax 806-742-4358 :
====================================================================
