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:20211207T055404Z
LOCATION:Online
DTSTART;TZID=America/Chicago:20211114T103000
DTEND;TZID=America/Chicago:20211114T110000
UID:submissions.supercomputing.org_SC21_sess430_ws_exampi102@linklings.com
SUMMARY:Leveraging Interconnect QoS Capabilities for Congestion-Aware MPI 
 Communication
DESCRIPTION:Workshop\n\nLeveraging Interconnect QoS Capabilities for Conge
 stion-Aware MPI Communication\n\nKhalilov, Slinka, Zhang\n\nResource shari
 ng between job allocations on production supercomputers often leads to tra
 ffic interference between MPI applications. Network congestion, a side eff
 ect of such resource sharing, results in substantial MPI latency degradati
 on, making application performance unpredictable. We consider the traffic 
 isolation capabilities of modern RDMA interconnects and propose a generic 
 Priority Assignment algorithm that associates a priority with each network
  Send operation using information about the network latency in the past. O
 ur implementation of Priority Assignment algorithm in UCX framework shows 
 up to the 22x GPCNeT benchmark improvement on 64 node cluster with InfiniB
 and EDR and OpenMPI in comparison to the reference UCX implementation. Pac
 ket-level simulation shows that proposed Priority Assignment algorithm hel
 ps to mitigate the effects of network congestion scaling up to the 512 nod
 es.\n\nTag: Online Only, Extreme Scale Computing, Parallel Programming Lan
 guages and Models, Performance\n\nRegistration Category: Workshop Reg Pass
END:VEVENT
END:VCALENDAR