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:20230124T171526Z
LOCATION:C144-145
DTSTART;TZID=America/Chicago:20221114T160000
DTEND;TZID=America/Chicago:20221114T161000
UID:submissions.supercomputing.org_SC22_sess462_ws_ai4s108@linklings.com
SUMMARY:Scalable Integration of Computational Physics Simulations with Mac
 hine Learning
DESCRIPTION:Workshop\n\nScalable Integration of Computational Physics Simu
 lations with Machine Learning\n\nBoyer, Brewer, Jude, Dettwiller\n\nIntegr
 ation of machine learning with simulation is part of a growing trend, howe
 ver, the augmentation of codes in a highly-performant, distributed manner 
 poses a software development challenge. In this work, we explore the quest
 ion of how to easily augment legacy simulation codes on high- performance 
 computers (HPCs) with machine-learned surrogate models, in a fast, scalabl
 e manner. Initial naive augmentation attempts required significant code mo
 dification and resulted in significant slowdown. This led us to explore in
 ference server techniques, which allow for model calls through drop-in fun
 ctions. In this work, we investigated TensorFlow Serving with gRPC and Red
 isAI with SmartRedis for server-client implementations, where the deep lea
 rning platform runs as a persistent process on HPC compute node GPUs and t
 he simulation makes client calls while running on CPUs. We evaluated infer
 ence performance for real gas equations of state, machine-learned boundary
  conditions for rotorcraft aerodynamics, and super-resolution techniques o
 n a POWER9 supercomputer.\n\nSession Format: Recorded\n\nRegistration Cate
 gory: Workshop Reg Pass
END:VEVENT
END:VCALENDAR