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:20230124T171522Z
LOCATION:C146
DTSTART;TZID=America/Chicago:20221118T092500
DTEND;TZID=America/Chicago:20221118T095500
UID:submissions.supercomputing.org_SC22_sess446_ws_ipdrm103@linklings.com
SUMMARY:SABO: Dynamic MPI+OpenMP Resource Balancer
DESCRIPTION:Workshop\n\nSABO: Dynamic MPI+OpenMP Resource Balancer\n\nRoch
 a Barbosa, Lemarinier, Papauré, Pérache, Krajecki\n\nScientific parallel a
 pplications often use MPI for inter-node communications and OpenMP for int
 ra-node orchestration. Parallel applications such as particle transport, s
 eismic wave propagation simulator, or finite-element applications often ex
 hibit workload imbalance due to their nature of ongoing data movement. The
 se applications usually develop software balancing strategies triggered wh
 en some imbalance thresholds are detected to reduce this imbalance. These 
 developments are complex to implement and impact the entire distributed ap
 plications’ performance by synchronizing and exchanging the load over the 
 network. This preseentation proposes a method to dynamically detect load i
 mbalance and balance the computation by redistributing OpenMP threads betw
 een MPI processes local to the node. With minimal impact on the applicatio
 ns’ codes, we demonstrate how this technique can improve the overall appli
 cations’ performance up to 28% on MiniFE, 17% on Quicksilver, and 3% on On
 des3D. We also present its impact when executing multiple nodes and our pr
 oposed approach’s limitations.\n\nSession Format: Recorded\n\nRegistration
  Category: Workshop Reg Pass
END:VEVENT
END:VCALENDAR