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:20260522T150110Z
LOCATION:C2/3/4 Ballroom
DTSTART;TZID=America/Chicago:20181114T083000
DTEND;TZID=America/Chicago:20181114T170000
UID:submissions.supercomputing.org_SC18_sess326_spost132@linklings.com
SUMMARY:Modeling Single-Source Shortest Path Algorithm Dynamics to Control
  Performance and Power Tradeoffs
DESCRIPTION:Sara Karamati, Jeffrey Young, and Rich Vuduc (Georgia Institut
 e of Technology)\n\nThis work presents a new methodology to improve the pe
 rformance of parallel algorithms by tuning the amount of available paralle
 lism for execution throughout the runtime. As such, we expose key paramete
 rs controlling the performance and parallelism of the algorithm and build 
 a software-based controller with the objective of maintaining the optimal 
 performance. Our controller allows for tuning the level of parallelism exe
 cuted in each time epoch to optimize for performance while preserving powe
 r usage. More specifically, our experimental evaluation focuses on a tunab
 le variation of a GPU-based delta-stepping algorithm for computing the sin
 gle-source shortest path (SSSP); As the available parallelism for the delt
 a-stepping SSSP is highly irregular and strongly input-dependent, our exte
 nsive experiments show that average power can be reduced while average par
 allelism is increased. This increase in average parallelism provides subst
 antial energy savings, independent of the hardware.\n\nRegistration Catego
 ry: Tech Program Reg Pass, Exhibits Reg Pass\n\n
END:VEVENT
END:VCALENDAR