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:20260522T150125Z
LOCATION:D161
DTSTART;TZID=America/Chicago:20181112T121000
DTEND;TZID=America/Chicago:20181112T123000
UID:submissions.supercomputing.org_SC18_sess158_ws_lasalss101@linklings.co
 m
SUMMARY:Non-Collective Scalable Global Network Based on Local Communicatio
 ns
DESCRIPTION:Marco Berghoff and Ivan Kondov (Karlsruhe Institute of Technol
 ogy)\n\nTo efficiently perform collective communications in current high-p
 erformance computing systems is a time-consuming task.\nWith future exasca
 le systems, this communication time will be increased further.\nHowever, g
 lobal information is frequently required in various physical models.\nBy e
 xploiting domain knowledge of the model behaviors globally needed informat
 ion can be distributed more efficiently, using only peer-to-peer communica
 tion which spread the information to all processes asynchronous during mul
 tiple communication steps.\nIn this article, we introduce a multi-hop base
 d Manhattan Street Network (MSN) for global information exchange and show 
 the conditions under which a local neighbor exchange is sufficient for exc
 hanging distributed information.\nBesides the MSN, in various models, glob
 al information is only needed in a spatially limited region inside the sim
 ulation domain.\nTherefore, a second network is introduced, the local exch
 ange network, to exploit this spatial assumption.\n\nBoth non-collective g
 lobal exchange networks are implemented in the massively parallel NAStJA f
 ramework.\nBased on two models, a phase-field model for droplet simulation
 s and the cellular Potts model for biological tissue simulations, we exemp
 lary demonstrate the wide applicability of these networks.\nScaling tests 
 of the networks demonstrate a nearly ideal scaling behavior with an effici
 ency of over 90%.\nTheoretical prediction of the communication time on fut
 ure exascale systems shows an enormous advantage of the presented exchange
  methods of O(1) by exploiting the domain knowledge.\n\nTag: Algorithms, H
 eterogeneous Systems, Resiliency\n\nRegistration Category: Workshop Reg Pa
 ss\n\nSession Chairs: Vassil Alexandrov (Hartree Centre, STFC); Jack Donga
 rra (University of Tennessee, Knoxville; Oak Ridge National Laboratory (OR
 NL)); Christian Engelmann (Oak Ridge National Laboratory (ORNL)); and Al G
 eist (Oak Ridge National Laboratory (ORNL))\n\n
END:VEVENT
END:VCALENDAR