Hi,
I don't think this is the case ... the new calculation bases the ERT on
the statistics about the waiting jobs in the queue. If the queue is
closed, I don't think you can put jobs in it, so there will be no
waiting jobs, so ERT will be low.
A quick check of
http://ui03.nikhef.nl:20987/
which has the current algorithm (the one submitted as an LCG patch and
scheduled to be included in 2.7.0) running behind it, shows this to
indeed be correct. qlong is closed and reports 60 seconds (the default)
for ERT.
I am of divided opinion about freeSlots == 0 when queue is closed.
You are right of course, but if the queue is closed, why is anyone
looking at freeSlots?
J "one advantage to having written something yourself" T
Tim Bell wrote:
> The ERT for a closed queue will continue to increase (based on the new
> calculations). The queue will therefore become less and less attractive.
> If the submitter wants to test for this condition, they can use the
> ComputeElement status value (production, closed, draining, etc.).
>
> FreeSlots should certainly be set to 0 if the queue is closed.
>
> Tim
|