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:20181221T160728Z
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:Workshop\nAlgorithms, Heterogeneous Systems, Resiliency, Works
 hop Reg Pass\n\nNon-Collective Scalable Global Network Based on Local Comm
 unications\n\nBerghoff, Kondov\n\nTo efficiently perform collective commun
 ications in current high-performance computing systems is a time-consuming
  task.\nWith future exascale systems, this communication time will be incr
 eased further.\nHowever, global information is frequently required in vari
 ous physical models.\nBy exploiting domain knowledge of the model behavior
 s globally needed information can be distributed more efficiently, using o
 nly peer-to-peer communication which spread the information to all process
 es asynchronous during multiple communication steps.\nIn this article, we 
 introduce a multi-hop based Manhattan Street Network (MSN) for global info
 rmation exchange and show the conditions under which a local neighbor exch
 ange is sufficient for exchanging distributed information.\nBesides the MS
 N, in various models, global information is only needed in a spatially lim
 ited region inside the simulation domain.\nTherefore, a second network is 
 introduced, the local exchange network, to exploit this spatial assumption
 .\n\nBoth non-collective global exchange networks are implemented in the m
 assively parallel NAStJA framework.\nBased on two models, a phase-field mo
 del for droplet simulations and the cellular Potts model for biological ti
 ssue simulations, we exemplary demonstrate the wide applicability of these
  networks.\nScaling tests of the networks demonstrate a nearly ideal scali
 ng behavior with an efficiency of over 90%.\nTheoretical prediction of the
  communication time on future exascale systems shows an enormous advantage
  of the presented exchange methods of O(1) by exploiting the domain knowle
 dge.
URL:https://sc18.supercomputing.org/presentation/?id=ws_lasalss101&sess=se
 ss158
END:VEVENT
END:VCALENDAR