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:605
DTSTART;TZID=America/Denver:20231113T153000
DTEND;TZID=America/Denver:20231113T155000
UID:submissions.supercomputing.org_SC23_sess453_ws_exampi108@linklings.com
SUMMARY:Optimizing Irregular Communication with Neighborhood Collectives a
 nd Locality-Aware Parallelism
DESCRIPTION:Gerald Collom (University of New Mexico), Rui Peng Li (Lawrenc
 e Livermore National Laboratory (LLNL)), and Amanda Bienz (University of N
 ew Mexico)\n\nIrregular communication limits both performance and scalabil
 ity of parallel applications. Typically, it is implemented as point-to-poi
 nt, and optimizations are integrated into the application, lacking portabi
 lity. Optimization of point-to-point messages within MPI is difficult, as 
 the interface only provides information on a piece of overall communicatio
 n. However, persistent neighbor collectives expose a suitable interface fo
 r such optimizations.\n\nThis paper presents methods for implementing exis
 ting optimizations for irregular communication within neighborhood collect
 ives, analyzes the impact of neighborhood collectives in Hypre BoomerAMG, 
 and shows up to a 1.38x speedup on sparse matrix-vector multiplication usi
 ng optimized neighbor collectives. The authors analyze three implementatio
 ns of neighborhood collectives for Alltoallv: an unoptimized wrapper of st
 andard point-to-point communication, and two locality-aware aggregating me
 thods. The second exposes a non-standard interface to perform additional o
 ptimization for an additional 0.07x speedup.\n\nOptimizations are availabl
 e open-source in MPI Advance which wraps MPI, allowing use with any MPI in
 stallation.\n\nTag: Exascale, Message Passing, Programming Frameworks and 
 System Software\n\nRegistration Category: Workshop Reg Pass\n\nSession Cha
 irs: Purushotham Bangalore (The University of Alabama); Amanda J. Bienz (U
 niversity of New Mexico); Matthew G. F. Dosanjh (Sandia National Laborator
 ies); Ryan Grant (Queen's University, Canada; Power API); William Schonbei
 n (Sandia National Laboratories); and Anthony Skjellum (Tennessee Technolo
 gical University, ASCEND-TNTECH)\n\n
END:VEVENT
END:VCALENDAR