Miron,
> It will help us when deciding how to "drain" nodes for jobs that
> require a multi-core machine or several nodes. It will be also used
> to run jobs in "back fill" mode in such cases. At this point we do
> not plan on using this infomration to assign priorities.
You can always run Standard Universe jobs up to the point that a node
is ready to be claimed for a single user regardless of any of these
constraints because they can be checkpointed within a few minutes.
I also don't see that a Standard Universe job really needs to specify
"estimated_run_hours" for that same reason.
I hope these policies are designed to take that into account.
--
Dan
|