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:20230831T095745Z
LOCATION:Seehorn
DTSTART;TZID=Europe/Stockholm:20230626T173000
DTEND;TZID=Europe/Stockholm:20230626T180000
UID:submissions.pasc-conference.org_PASC23_sess161_msa183@linklings.com
SUMMARY:Scalable GPU-Accelerated Incremental Checkpointing of Sparsely Upd
 ated Data
DESCRIPTION:Minisymposium\n\nMichela Taufer and Nigel Phillip Tan (Univers
 ity of Tennessee), Bogdan Nicolae (Argonne National Laboratory), Jakob Lue
 ttgau and Sanjukta Bhowmick (University of Tennessee), Keita Teranishi (Oa
 k Ridge National Laboratory), Nicolas Morales (Sandia National Laboratorie
 s), and Franck Cappello (Argonne National Laboratory)\n\nCheckpointing lar
 ge amounts of related data concurrently to stable storage is a common I/O 
 pattern of many HPC applications in various scenarios including checkpoint
 -restart fault tolerance, coupled workflows that combine simulations with 
 analytics, and adjoint computations. This pattern is challenging because i
 t needs to happen frequently and typically leads to I/O bottlenecks that n
 egatively impact the performance and scalability of the applications. A la
 rge class of applications including graph algorithms performs sparse updat
 es between checkpoints. Incremental checkpointing approaches that save onl
 y the differences from one checkpoint to another can dramatically reduce I
 /O bottlenecks and storage utilization. However, such techniques are not w
 ithout challenges: it is non-trivial to determine what data changed since 
 a previous checkpoint transparently and to assemble the differences in a c
 ompact fashion that does not result in excessive metadata. This talk discu
 sses the challenge of making efficient incremental checkpoints on GPU-acce
 lerated platforms and introduces an innovative approach that builds a comp
 act representation of the differences between checkpoints using Merkle-tre
 e-inspired data structures for parallel data construction and manipulation
 . We assess the effectiveness of our approach with ORANGES, a graph alignm
 ent application with sparse update patterns.\n\nDomain: Computer Science, 
 Machine Learning, and Applied Mathematics 
\n\nSession Chair: Nicolas Mora
 les (Sandia National Laboratories)
END:VEVENT
END:VCALENDAR