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:20230124T171523Z
LOCATION:C1-2-3
DTSTART;TZID=America/Chicago:20221115T083000
DTEND;TZID=America/Chicago:20221115T170000
UID:submissions.supercomputing.org_SC22_sess250_drs101@linklings.com
SUMMARY:Load Balancing, Fault Tolerance, and Resource Elasticity for Async
 hronous Many-Task (AMT) Systems
DESCRIPTION:Doctoral Showcase, Posters\n\nLoad Balancing, Fault Tolerance,
  and Resource Elasticity for Asynchronous Many-Task (AMT) Systems\n\nPosne
 r\n\nTo enable efficient and productive programming of today's supercomput
 ers and beyond, a variety of issues must be addressed, including: load bal
 ancing (i.e., utilizing all resources equally), fault tolerance (i.e., cop
 ing with hardware failures), and resource elasticity (i.e., allowing the a
 ddition/release of resources).\n\nIn this work, we address above issues in
  the context of Asynchronous Many-Tasking (AMT) for clusters. Here, progra
 mmers split a computation into many fine-grained execution units (called t
 asks), which are dynamically mapped to processing units (called workers) b
 y a runtime system. \n\nRegarding load balancing, we propose a work steali
 ng technique that transparently schedules tasks to resources of the overal
 l system, balancing the workload over all processing units. Experiments sh
 ow good scalability, and a productivity evaluation shows intuitive use.\n\
 nRegarding fault tolerance, we propose four techniques to protect programs
  transparently. All perform localized recovery and continue the program ex
 ecution with fewer resources. Three techniques write uncoordinated checkpo
 ints of task descriptors in a resilient store. One technique does not writ
 e checkpoints, but exploits natural task duplication of work stealing. Exp
 eriments show failure-free running time overhead below 1% and a recovery o
 verhead below 0.5 seconds. Simulations of job set executions show that mak
 espans can be reduced by up to 97%.\n\nRegarding resource elasticity, we p
 ropose a technique to enable the addition and release of nodes at runtime 
 by transparently relocating tasks accordingly. Experiments show costs for 
 adding and releasing nodes below 0.5 seconds. Additionally, simulations of
  job set executions show that makespans can be reduced by up to 20%.\n\nRe
 gistration Category: Tech Program Reg Pass, Exhibits Reg Pass
END:VEVENT
END:VCALENDAR