BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260422T143140Z
LOCATION:B314
DTSTART;TZID=America/New_York:20241117T103000
DTEND;TZID=America/New_York:20241117T110000
UID:submissions.supercomputing.org_SC24_sess741_ws_mlg102@linklings.com
SUMMARY:MDLoader: A Hybrid Model-Driven Data Loader for Distributed Graph 
 Neural Network Training
DESCRIPTION:Jonghyun Bae (Google); Jong Youl Choi, Massimiliano Lupo Pasin
 i, Kshitij Mehta, and Pei Zhang (Oak Ridge National Laboratory (ORNL)); an
 d Khaled Z. Ibrahim (Lawrence Berkeley National Laboratory (LBNL))\n\nScal
 able data management is essential for processing large scientific dataset 
 on HPC platforms for distributed deep learning. In-memory distributed stor
 age is preferred for its speed, enabling rapid, random, and frequent data 
 access required by stochastic optimizers. Processes use one-sided or colle
 ctive communication to fetch remote data, with optimal performance dependi
 ng on (i) dataset characteristics, (ii) training scale, and (iii) intercon
 nection network. Empirical analysis shows collective communication excels 
 with larger mini-batch sizes and/or fewer processes, whereas one-sided com
 munication outperforms at larger scales.\n\nWe propose MDLoader, a hybrid 
 in-memory data loader for distributed graph neural network training. MDLoa
 der features a model-driven performance estimator that dynamically selects
  between one-sided and collective communication at the beginning of traini
 ng using Tree of Parzen Estimators (TPE). Evaluations on NERSC Perlmutter 
 and OLCF Summit show MDLoader outperforms single-backend loaders by up to 
 2.83x and predicts the suitable communication method with 96.3% (Perlmutte
 r) and 94.3% (Summit) success rate.\n\nTag: Artificial Intelligence/Machin
 e Learning, Graph Algorithms, Scalable Data Mining\n\nRegistration Categor
 y: Workshop Reg Pass\n\nSession Chairs: Seung-Hwan Lim (Oak Ridge National
  Laboratory (ORNL)); José Moreira (IBM); Catherine Schuman (University of 
 Tennessee, Knoxville); and Richard Vuduc (Georgia Institute of Technology)
 \n\n
END:VEVENT
END:VCALENDAR