BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T000712Z
LOCATION:401-402
DTSTART;TZID=America/Denver:20231114T133000
DTEND;TZID=America/Denver:20231114T140000
UID:submissions.supercomputing.org_SC23_sess180_pap191@linklings.com
SUMMARY:Mirage: Toward Low-interruption Services on Batch GPU Clusters wit
 h Reinforcement Learning
DESCRIPTION:Qiyang Ding (University of Texas), Pengfei Zheng (University o
 f Wisconsin), Shreyas Kudari (University of Texas), Shivaram Venkataraman 
 (University of Wisconsin), and Zhao Zhang (Texas Advanced Computing Center
  (TACC))\n\nAccommodating long-running deep learning (DL) training and inf
 erence jobs is challenging on GPU clusters that use traditional batch sche
 dulers, such as Slurm. Given fixed wall clock time limits, DL researchers 
 usually need to run a sequence of batch jobs and experience long interrupt
 ions on overloaded machines. Such interruptions significantly lower the re
 search productivity and QoS for services that are deployed in production. 
 To mitigate the issues from interruption, we explore a set of machine lear
 ning and reinforcement learning techniques to design a proactive provision
 er. We examine the generality of the method using production job traces fr
 om three GPU clusters.  We validate the effectiveness and generality of ou
 r proactive provisioner using the validation trace of each cluster. Our ex
 periments show that the proposed resource provisioner safeguards 23%-76% o
 f jobs with zero interruption across varying load levels on the three clus
 ters.\n\nTag: Architecture and Networks, Performance Measurement, Modeling
 , and Tools, Resource Management\n\nRegistration Category: Tech Program Re
 g Pass\n\nReproducibility Badges: Artifact Available, Artifact Functional,
  Results Reproduced\n\nSession Chair: Ann Gentile (Sandia National Laborat
 ories)\n\n
END:VEVENT
END:VCALENDAR