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:20260422T000711Z
LOCATION:401-402
DTSTART;TZID=America/Denver:20231114T143000
DTEND;TZID=America/Denver:20231114T150000
UID:submissions.supercomputing.org_SC23_sess180_pap474@linklings.com
SUMMARY:DPS: Adaptive Power Management for Overprovisioned Systems
DESCRIPTION:Jianru Ding and Henry Hoffmann (University of Chicago)\n\nMaxi
 mizing performance under a power budget is essential for HPC systems and h
 as inspired the development of many power management frameworks.  These ca
 n be broadly characterized into two groups: model-based and stateless.  Mo
 del-based frameworks achieve good performance under a power budget but are
  highly dependent on the quality of the model and the data used to train i
 t. Stateless frameworks are more robust and require no training, but are g
 enerally lower performance.  In this paper, we propose a new framework tha
 t does not require a model, but does track state in the form of recent pow
 er dynamics. We implement this idea and test it on a public cloud running 
 both Spark and HPC jobs.  We find when total power demand is low, our fram
 ework achieves equivalent performance to prior work, but when power demand
  is high it achieves a mean 8% performance improvement (with no reliance o
 n a learned model).\n\nTag: Architecture and Networks, Performance Measure
 ment, Modeling, and Tools, Resource Management\n\nRegistration Category: T
 ech Program Reg Pass\n\nReproducibility Badges: Artifact Available, Artifa
 ct Functional, Results Reproduced\n\nSession Chair: Ann Gentile (Sandia Na
 tional Laboratories)\n\n
END:VEVENT
END:VCALENDAR