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:20260202T201803Z
LOCATION:242
DTSTART;TZID=America/Chicago:20251116T101000
DTEND;TZID=America/Chicago:20251116T103500
UID:submissions.supercomputing.org_SC25_sess216_ws_isav103@linklings.com
SUMMARY:In-Transit Data Transport Strategies for Coupled AI-Simulation Wor
 kflow Patterns
DESCRIPTION:Harikrishna Tummalapalli, Riccardo Balin, and Christine M. Sim
 pson (Argonne National Laboratory (ANL)); Andrew Park and Aymen Alsaadi (R
 utgers University); Andrew E. Shao (Hewlett Packard Enterprise (HPE)); Wes
 ley Brewer (Oak Ridge National Laboratory (ORNL)); and Shantenu Jha (Rutge
 rs University)\n\nCoupled AI-Simulation workflows are becoming the major w
 orkloads for HPC facilities, and their increasing complexity necessitates 
 new tools for performance analysis and prototyping of new in-situ workflow
 s. We present SimAI-Bench, a tool designed to both prototype and evaluate 
 these coupled workflows. In this paper, we use SimAI-Bench to benchmark th
 e data transport performance of two common patterns on the Aurora supercom
 puter: a one-to-one workflow with co-located simulation and AI training in
 stances, and a many-to-one workflow where a single AI model is trained fro
 m an ensemble of simulations. For the one-to-one pattern, our analysis sho
 ws that node-local and DragonHPC data staging strategies provide excellent
  performance compared Redis and Lustre file system. For the many-to-one pa
 ttern, we find that data transport becomes a dominant bottleneck as the en
 semble size grows. Our evaluation reveals that file system is the optimal 
 solution among the tested strategies for the many-to-one pattern.\n\nTag: 
 AI, Machine Learning, & Deep Learning, Clouds & Distributed Computing, Per
 formance Evaluation, Scalability, & Portability, Scientific & Information 
 Visualization\n\nRecording: Partially Livestreamed, Partially Recorded\n\n
 Registration Category: Technical Program Reg Pass, Workshop Reg Pass\n\nSe
 ssion Chairs: Axel Huebl (Lawrence Berkeley National Laboratory (LBNL)); W
 ill Usher (Luminary Cloud Inc); Nicola Ferrier (Argonne National Laborator
 y (ANL), University of Chicago); Estelle Dirand (TotalEnergies, France); V
 ictor Mateevitsi (Argonne National Laboratory (ANL), University of Illinoi
 s Chicago); and Jens Henrik Göbbert (Forschungszentrum Jülich, Jülich Supe
 rcomputing Centre (JSC))\n\n
END:VEVENT
END:VCALENDAR