Food for thought.
I've been following this thread with some interest because
I had been wondering how best to handle SGI's machines
in a remote beaker lab.
There are a few fundamental problems:
1) The machines are sometimes unique -- only one exists, period.
The result is that at certain times the machines must be available
to particular users -- customer/trade show demos for example.
2) They often require hours of continuous time which, particularly
during bring up and major OS releases means scheduling
evenings and weekends --- engineers are arranging their sleep
schedules to take advantage of certain time slots. Random
start times for reservations won't work.
What seems to be lacking in beaker is *Hard Start Times* which
implies hard-end-times.
So for example, if we put a 4 rack UV system (2048 cores, 16Tb of
memory) in beaker it is not just going to stay there; it will be pulled
back inside SGI's firewall on a regular basis.
I was hoping that reservations would address this issue.
Thoughts?
George