​When you submit jobs and use the --memory option you can give units in both MB and GB. jobsub_submit interprets 1 GB as 1024 MB, not 1000 MB. So --memory=2GB is equivalent to --memory=2048MB, not 2000MB.


​You are more likely to get a grid slot for your job if you request the lifetime you actually need. This will be enforced in early summer 2016.


​Are priority numbers comparable across experiments?

  • If two different exps have same priority (a tie) – there is a tie-breaking mechanism. Otherwise are the same.

​How are resources determined between experiments? Jobs running at a given moment – tune priority factors, and relaxation factor)

  • Priority factor of accounting group: per user
    • How many jobs are running, how many running recently (past 1.5 days approx.)
    • Only changed when user asks.  Default to start with is same for all.
  • Relaxation factor : is global

​How do priorities affect grid job startup?

  • Lower priority numbers are better.  Assuming resource matching. Lowest priority wins. Exponential relaxation time.

​Jobs that run past their resource limits get held, not killed (do jobsub_q to see reason why). Understand why, kill jobs and resubmit.