BEGIN:VCALENDAR VERSION:2.0 PRODID:Linklings LLC BEGIN:VTIMEZONE TZID:America/New_York X-LIC-LOCATION:America/New_York BEGIN:DAYLIGHT TZOFFSETFROM:-0500 TZOFFSETTO:-0400 TZNAME:EDT DTSTART:19700308T020000 RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU END:DAYLIGHT BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 TZNAME:EST DTSTART:19701101T020000 RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20210402T160051Z LOCATION:Track 3 DTSTART;TZID=America/New_York:20201117T150000 DTEND;TZID=America/New_York:20201117T153000 UID:submissions.supercomputing.org_SC20_sess155_pap573@linklings.com SUMMARY:RLScheduler: An Automated HPC Batch Job Scheduler Using Reinforcem ent Learning DESCRIPTION:Paper\n\nRLScheduler: An Automated HPC Batch Job Scheduler Usi ng Reinforcement Learning\n\nZhang, Dai, He, Bao, Xie\n\nToday's high-perf ormance computing platforms are still dominated by batch jobs. Accordingly , effective batch job scheduling is crucial to obtain high system efficien cy. Existing batch job schedulers typically leverage heuristic priority fu nctions to prioritize and schedule jobs. Once configured by the experts, s uch priority functions can hardly adapt to the changes of job loads, optim ization goals or system settings, potentially leading to degraded system e fficiency when changes occur. To address this fundamental issue, we presen t RLScheduler, an automated HPC batch job scheduler built on reinforcement learning. RLScheduler relies on minimal manual interventions or expert kn owledge, but can learn high-quality scheduling policies via its own contin uous 'trial and error'. Through extensive evaluations, we confirm that RLS cheduler can learn high-quality scheduling policies towards various worklo ads and optimization goals with relatively low computation cost. Moreover, we show that the learned models perform stably even applied to unseen wor kloads, making them practical for production use.\n\nTag: File Systems and I/O, Machine Learning, Deep Learning and Artificial Intelligence, Perform ance/Productivity Measurement and Evaluation, Resource Management and Sche duling\n\nRegistration Category: Tech Program Reg Pass END:VEVENT END:VCALENDAR