BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//juliacon-2026//speaker//WXQ9M8
BEGIN:VTIMEZONE
TZID:Europe/Berlin
BEGIN:DAYLIGHT
DTSTART:20250812T000000
TZNAME:CEST
TZOFFSETFROM:+0200
TZOFFSETTO:+0200
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20251026T030000
RDATE:20261025T030000
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20260329T030000
RDATE:20270328T030000
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
SUMMARY:Parallel Processing in JetReconstruction.jl - Harshit Nagpal\, Gra
 eme Stewart\, Mateusz Fila
DTSTART;TZID=Europe/Berlin:20260812T103000
DTEND;TZID=Europe/Berlin:20260812T110000
DTSTAMP:20261009T181015Z
UID:pretalx-juliacon-2026-F3FBUY@pretalx.com
DESCRIPTION:The Julia JetReconstruction.jl package was released in 2024 an
 d provides a\nhigh-performance native Julia implementation of serial jet r
 econstruction in\nhigh-energy physics\, inspired by the FastJet C++ packag
 e. In serial mode the\ncode outperforms FastJet by 14-40%\, depending on t
 he exact parameters used.\n\nUntil now\, no systematic testing and benchma
 rking of the code had been done\nwhen running in a parallel\, multi-thread
 ed\, mode. In this paper we present the\nfirst results of doing this. The 
 initial parallel performance of\nJetReconstruction.jl\, compared to an equ
 ivalent multi-threaded version of\nFastJet\, showed increasingly poor perf
 ormance\, with the advantage of\nJetReconstruction.jl quickly evaporating 
 and becoming about a x10 deficit at\nhigh thread counts in the worst case.
 \n\nWe performed an investigation of the underlying reasons for this poor
 \nperformance\, looking at thread scheduling parameters\, physics code sca
 ling\, and\nmemory allocations and garbage collection issues. We review th
 is performance\ninvestigation\, and the profiling tools which we used\, an
 d how we finally\nunderstood the source of the problem: memory allocations
  and garbage collection.\n\nWe have now designed a new approach to reuse a
 lready allocated memory\, with\nmuch better parallel scaling. We discuss e
 rgonomics of this new interface\, which\nnow allows users to run in a high
 ly parallel mode with excellent performance\,\ntested up to 64 parallel th
 reads\, which consistently maintains the advantage of\nJetReconstruction.j
 l over FastJet.
LOCATION:Alte Mensa — Audi Max
URL:https://pretalx.com/juliacon-2026/talk/F3FBUY/
END:VEVENT
END:VCALENDAR