BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20211207T054814Z
LOCATION:230-231-232
DTSTART;TZID=America/Chicago:20211118T160000
DTEND;TZID=America/Chicago:20211118T163000
UID:submissions.supercomputing.org_SC21_sess167_pap609@linklings.com
SUMMARY:Online Evolutionary Batch Size Orchestration for Scheduling Deep L
 earning Workloads in GPU Clusters
DESCRIPTION:Paper\n\nOnline Evolutionary Batch Size Orchestration for Sche
 duling Deep Learning Workloads in GPU Clusters\n\nBian, Li, Wang, You\n\nE
 fficient GPU resource scheduling is essential to maximize resource utiliza
 tion and save training costs for the increasing amount of deep learning wo
 rkloads in shared GPU clusters. Existing GPU schedulers largely rely on st
 atic policies to leverage the performance characteristics of deep learning
  jobs. They can hardly reach optimal efficiency, however, due to the lack 
 of elasticity. To address the problem, we propose ONES, an ONline Evolutio
 nary Scheduler for elastic batch size orchestration. ONES automatically ma
 nages the elasticity of each job based on the training batch size, so as t
 o maximize GPU utilization and improve scheduling efficiency. It determine
 s the batch size for each job through an online evolutionary search that c
 an continuously optimize the scheduling decisions. We evaluate the effecti
 veness of ONES with 64 GPUs on TACC's Longhorn supercomputers. The results
  show that ONES can outperform the prior deep learning schedulers with a s
 ignificantly shorter average job completion time.\n\nTag: Reproducibility 
 Badge, Accelerator-based Architectures, Resource Management and Scheduling
 \n\nRegistration Category: Tech Program Reg Pass\n\nReproducibility Badges
 : Artifact Available, Artifact Functional, Results Reproduced
END:VEVENT
END:VCALENDAR