BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260421T090514Z
LOCATION:Bldg. 6 - Room 103
DTSTART;TZID=Europe/Stockholm:20260630T144500
DTEND;TZID=Europe/Stockholm:20260630T151500
UID:submissions.pasc-conference.org_PASC26_sess145_msa179@linklings.com
SUMMARY:AI-Enabled Modeling, Simulation, and Optimization of Distributed C
 omputing Systems
DESCRIPTION:Sairam Sri Vatsavai (Brookhaven National Laboratory)\n\nDistri
 buted computing infrastructures are growing in complexity and heterogeneit
 y, challenging traditional modeling and simulation methodologies. Analytic
 al approaches such as queueing-theoretic and Markov chain models, while ma
 thematically tractable, rely on simplifying assumptions that fail to captu
 re the stochastic dynamics of large-scale systems. Discrete-event simulati
 on offers higher fidelity but scales poorly with infrastructure size, crea
 ting a fundamental trade-off between fidelity and speed. For federated sys
 tems that span hundreds of distributed sites, simulation-based evaluation 
 of new workload and data movement policies becomes prohibitively expensive
 , while production experimentation remains impractical. AI-enabled modelin
 g and simulation offers a promising path forward.\nThis talk presents our 
 work on AI-enabled modeling and simulation of distributed computing system
 s. We introduce a transformer-based surrogate trained on discrete-event si
 mulation traces that emulates job lifecycle dynamics including scheduling,
  resource contention, and data transfer, enabling orders-of-magnitude fast
 er evaluation. We discuss how simulation engines such as SimGrid can gener
 ate high-fidelity training datasets. The resulting surrogate serves as a d
 igital twin, enabling agentic AI to rapidly explore diverse workload and d
 ata allocation policies to improve resource utilization and resilience, wh
 ile supporting decision-making on infrastructure changes by quantifying th
 eir impact on system performance. These capabilities establish a framework
  for distributed computing systems co-design.\n\nDomain: Engineering, Phys
 ics, Computational Methods and Applied Mathematics\n\nSession Chairs: Raee
 s Ahmad Khan (University of Pittsburgh, CERN); Tatiana Korchuganova (Unive
 rsity of Pittsburgh, CERN); and Alexei Klimentov (Brookhaven National Labo
 ratory)\n\n
END:VEVENT
END:VCALENDAR