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:20210402T160048Z
LOCATION:Track 3
DTSTART;TZID=America/New_York:20201117T130000
DTEND;TZID=America/New_York:20201117T133000
UID:submissions.supercomputing.org_SC20_sess141_pap142@linklings.com
SUMMARY:ScalAna: Automating Scaling Loss Detection with Graph Analysis
DESCRIPTION:Paper\n\nScalAna: Automating Scaling Loss Detection with Graph
  Analysis\n\nJin, Wang, Yu, Tang, Hoefler...\n\nScaling a parallel program
  to modern supercomputers is challenging due to inter-process communicatio
 n, Amdahl's law and resource contention. Performance analysis tools for fi
 nding such scaling bottlenecks are based on either profiling or tracing. P
 rofiling incurs low overheads but does not capture detailed dependencies n
 eeded for root-cause analysis. Tracing collects all information at prohibi
 tive overheads.\n\nIn this work, we design ScalAna that uses static analys
 is techniques to enable the analyzability of traces at a cost similar to p
 rofiling. ScalAna first leverages static compiler techniques and runtime l
 ightweight techniques to build a Program Performance Graph. With this grap
 h, we propose a novel backtracking algorithm to automatically detect the r
 oot causes. We evaluate ScalAna with real applications. Results show that 
 ScalAna can effectively locate the root causes and incurs 1.73% overhead o
 n average for up to 2048 processes. We achieve up to 11.11% performance im
 provement on 2048 processes by fixing the root causes.\n\nTag: Graph Algor
 ithms, Performance/Productivity Measurement and Evaluation\n\nRegistration
  Category: Tech Program Reg Pass
END:VEVENT
END:VCALENDAR