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:20211207T055402Z
LOCATION:Second Floor Atrium
DTSTART;TZID=America/Chicago:20211116T083000
DTEND;TZID=America/Chicago:20211116T170000
UID:submissions.supercomputing.org_SC21_sess256_drs101@linklings.com
SUMMARY:Accelerating I/O for Traditional HPC and Modern ML Workloads on Em
 erging HPC Systems
DESCRIPTION:Doctoral Showcase, Posters\n\nAccelerating I/O for Traditional
  HPC and Modern ML Workloads on Emerging HPC Systems\n\nChien, Markidis, P
 odobas\n\nIn recent years, HPC systems have emerged as an attractive optio
 n to speed up large-scale Machine Learning (ML) workloads. HPC systems can
  tremendously improve learning speed with lots of GPUs and fast interconne
 ct. However, ML workloads that are data-intensive differ significantly fro
 m traditional HPC I/O. Popular ML frameworks are also not optimized for HP
 C hardware. At the same time, the I/O subsystems in emerging HPC machines 
 are becoming increasingly heterogeneous (e.g. object storage, NVMe) and di
 saggregated (node-local storage). It is unclear what is the most efficient
  way to leverage these emerging I/O systems for both traditional HPC and e
 merging ML workloads. In this thesis, we tackle the challenges from two di
 rections. Firstly, we explore the challenges of I/O when running emerging 
 ML workloads on existing HPC hardware. To achieve this, we research and de
 velop profiling tools that are coupled with ML workloads. We illustrate ho
 w information from our tools is invaluable for I/O performance tuning and 
 the importance of co-designing I/O with underlying workloads. Secondly, we
  explore emerging I/O subsystems. In particular, we focus on a disruptive 
 solution called object storage. While widely used in cloud applications, i
 t is still unclear what is a suitable programming model for HPC applicatio
 ns. We develop an object store emulator that supports parallel I/O to show
  that it can tremendously improve I/O bandwidth comparing to shared-file c
 ollective I/O. Finally, to conclude this work, we present our work-in-prog
 ress programming models for writing parallel shared-file through disaggreg
 ated fast local storage and a near-storage ML preprocessing accelerator.\n
 \nTag: In-Person Only\n\nRegistration Category: Tech Program Reg Pass, Exh
 ibit Hall Only
END:VEVENT
END:VCALENDAR