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:275
DTSTART;TZID=America/Chicago:20251119T133000
DTEND;TZID=America/Chicago:20251119T135200
UID:submissions.supercomputing.org_SC25_sess283_pap863@linklings.com
SUMMARY:STELLAR: Storage Tuning Engine Leveraging LLM Autonomous Reasoning
  for High-Performance Parallel File Systems
DESCRIPTION:Chris Egersdoerfer (University of Delaware); Philip Carns, Sha
 ne Snyder, and Robert Ross (Argonne National Laboratory (ANL)); and Dong D
 ai (University of Delaware)\n\nI/O performance is crucial to efficiency in
  data-intensive scientific computing, but tuning large-scale storage syste
 ms is complex, costly, and notoriously manpower-intensive, making it inacc
 essible for most domain scientists. In this study, we propose STELLAR, an 
 autonomous tuner for high-performance parallel file systems. Our evaluatio
 ns show that STELLAR always selects near-optimal configurations for the pa
 rallel file systems within the first five attempts, even for previously un
 seen applications. STELLAR’s human-like efficiency is fundamentally differ
 ent from existing auto-tuning methods, which often require hundreds of tho
 usands of iterations to converge. STELLAR achieves this through Retrieval-
 Augmented Generation, external tool execution, LLM-based reasoning, and a 
 multi-step agent design to stabilize reasoning and combat hallucinations. 
 STELLAR's architecture opens new avenues for addressing complex system opt
 imization problems, especially those characterized by vast search spaces a
 nd high exploration costs. Its extremely efficient autonomous tuning will 
 broaden access to I/O performance optimizations for domain scientists with
  minimal additional resource investment.\n\nTag: Data Analytics, Visualiza
 tion & Storage\n\nRecording: Livestreamed, Recorded\n\nRegistration Catego
 ry: Technical Program Reg Pass\n\nSession Chair: Ana Kupresanian (Lawrence
  Berkeley National Laboratory (LBNL))\n\n
END:VEVENT
END:VCALENDAR