LCLS Needs

• On-shift: jobs start in ~1 minute
• Off-shift: jobs start "soon"
• Normal analysis: standard non-killable jobs (LCLS doesn't checkpoint)

Previous implementations:

• at SLAC: multiple levels of suspend-preemption
• at NERSC: reservations

On-shift Options

• reservations (wasteful)
• kill-preemption (hard on users, less so if jobs can be automatically resubmitted?)
• suspend-preemption (reduces available memory, difficult in a shared environment like S3DF)

Off-shift Options

• move-to-head-of-queue QOS
• suspend or kill preemption