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:20260202T201804Z
LOCATION:264
DTSTART;TZID=America/Chicago:20251117T164200
DTEND;TZID=America/Chicago:20251117T170000
UID:submissions.supercomputing.org_SC25_sess208_ws_worksp118@linklings.com
SUMMARY:Evaluating the Efficacy of LLM-Based Reasoning for Multiobjective 
 HPC Job Scheduling
DESCRIPTION:Prachi Jadhav (University of Tennessee, Knoxville; Oak Ridge N
 ational Laboratory (ORNL)); Hongwei Jin (Argonne National Laboratory (ANL)
 ); Ewa Deelman (University of Southern California (USC)); and Prasanna Bal
 aprakash (Oak Ridge National Laboratory (ORNL))\n\nHigh-Performance Comput
 ing job scheduling involves balancing conflicting objectives such as minim
 izing makespan, reducing wait times, optimizing resource use, and ensuring
  fairness. Heuristic-based methods, e.g., FJFS and SJF or intensive optimi
 zation techniques, often lack adaptability to dynamic workloads and cannot
  simultaneously optimize multiobjectives in HPC systems. We propose a nove
 l LLM-based scheduler using a ReAct-style framework, enabling iterative, i
 nterpretable decision-making. It incorporates a scratchpad memory to track
  scheduling history and refine decisions via natural language feedback, wh
 ile a constraint enforcement module ensures feasibility and safety.\nWe ev
 aluate our approach using OpenAI's O4-Mini and Anthropic's Claude 3.7 acro
 ss seven workload scenarios; heterogeneous mixes, bursty patterns, etc. Th
 e comparisons reveals that LLM-based scheduling effectively balances multi
 ple objectives while offering transparent reasoning through natural langua
 ge traces. The method excels in constraint satisfaction and adapts to dive
 rse workloads without domain-specific training. However, a trade-off betwe
 en reasoning quality and computational overhead challenges real-time deplo
 yment.\n\nRecording: Livestreamed, Recorded\n\nRegistration Category: Tech
 nical Program Reg Pass, Workshop Reg Pass\n\nSession Chairs: Silvina Caino
 -Lores (National Institute for Research in Digital Science and Technology 
 (Inria)) and Anirban Mandal (Renaissance Computing Institute (RENCI), Univ
 ersity of North Carolina at Chapel Hill)\n\n
END:VEVENT
END:VCALENDAR