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:20260624T171340Z
LOCATION:Bldg. 8 - Entrance Hall
DTSTART;TZID=Europe/Stockholm:20260630T173000
DTEND;TZID=Europe/Stockholm:20260630T194500
UID:submissions.pasc-conference.org_PASC26_sess135_pos131@linklings.com
SUMMARY:P33 - RDQ: A Zero-Copy Remote Data Queue for In-Situ Machine Learn
 ing in HPC
DESCRIPTION:Maximilian Sander (TU Dresden) and Jens Domke (RIKEN)\n\nThe i
 ntegration of machine learning into the computational sciences is increasi
 ngly pursued to reduce time-to-solution, alleviate I/O bottlenecks, and en
 able adaptive analysis during simulation. We present RDQ (Remote Data Queu
 e), a library for coupling HPC simulations and machine learning training u
 sing an MPMD (Multiple Program Multiple Data) MPI (Message Passing Interfa
 ce) approach. RDQ is part of an ongoing effort to combine HPC and ML workf
 lows and serves as a research platform for investigating in-situ training 
 and data reduction techniques. RDQ extends existing data staging mechanism
 s by enabling zero-copy tensor concatenation and put-side data sampling fo
 r in-situ learning scenarios. RDQ is based on MPI-3 RMA operations and req
 uires no user-managed compute resources, allowing flexible placement of da
 ta queues within the communicator.\nInternally, RDQ employs a multi–ring b
 uffer design that supports zero-copy receives, including direct access fro
 m Python and PyTorch via the DLPack tensor interface. Performance measurem
 ents show that RDQ achieves up to 2600 messages per second for small messa
 ges and saturates a 100 Gb/s network interface with message sizes of 8 MiB
 .\n\n
END:VEVENT
END:VCALENDAR