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:20210402T160051Z
LOCATION:Track 5
DTSTART;TZID=America/New_York:20201117T150000
DTEND;TZID=America/New_York:20201117T153000
UID:submissions.supercomputing.org_SC20_sess161_pap540@linklings.com
SUMMARY:A Hierarchical and Load-Aware Design for Large Message Neighborhoo
 d Collectives
DESCRIPTION:Paper\n\nA Hierarchical and Load-Aware Design for Large Messag
 e Neighborhood Collectives\n\nGhazimirsaeed, Zhou, Ruhela, Bayatpour, Subr
 amoni...\n\nThe MPI-3.0 standard introduced neighborhood collective to sup
 port sparse communication patterns used in many applications.  In this pap
 er, we propose a hierarchical and distributed graph topology that consider
 s the physical topology of the system and the virtual communication patter
 n of processes to improve the performance of large message neighborhood co
 llectives. Moreover, we propose two design alternatives on top of the hier
 archical design:  1) LAG-H, which assumes the same communication load for 
 all processes; and 2) LAW-H, which considers the communication load of pro
 cesses for fair distribution of load among them. We propose a mathematical
  model to determine the communication capacity of each process, then use t
 he derived capacity to fairly distribute the load among processes. Our exp
 erimental results on up to 28,672 processes show up to 9x speedup for vari
 ous process topologies. We also observe up to 8.2% performance gain and 34
 x speedup for NAS-DT and SpMM application kernels, respectively.\n\nTag: A
 lgorithms, MPI, Networks\n\nRegistration Category: Tech Program Reg Pass
END:VEVENT
END:VCALENDAR