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:20210916T132447Z
LOCATION:Mère Royaume
DTSTART;TZID=Europe/Stockholm:20210705T153000
DTEND;TZID=Europe/Stockholm:20210705T160000
UID:submissions.pasc-conference.org_PASC21_sess198_msa186@linklings.com
SUMMARY:From Sharing Work to Sharing Tasks: A Scheduling Perspective
DESCRIPTION:Minisymposium\n\nFrom Sharing Work to Sharing Tasks: A Schedul
 ing Perspective\n\nMohammed, Cavelan, Ciorba\n\nScientific applications ar
 e the cornerstone of computational sciences. They are considered a tool fo
 r insight for domain scientists. These applications are oftentimes large, 
 computationally-intensive, and highly parallel applications, which are usu
 ally executed on high performance computing (HPC) systems. Applications ne
 ed to fully exploit the hardware parallelism in the HPC systems to harness
  their computational power.  Parallelism in applications is usually e
 xposed and expressed in the form of data-parallel loops or tasks, that are
  independent, i.e., embarrassingly parallel, or may have data dependencies
 . Load imbalance, defined as uneven finishing times of processing elements
  (PEs), often degrades parallel applications performance and scalability. 
 Load imbalance is caused by application, problem, and/or systemic characte
 ristics. Dynamic and adaptive scheduling addresses load imbalance in scien
 tific applications and improves their performance and scalability.  W
 hile dynamic loop scheduling is well understood and extensively experiment
 ed with using self‑scheduling techniques in data-parallel multithrea
 ded applications, adaptive and dynamic self-scheduling for task-parallel a
 pplications has not yet been studied and tested. In this talk, we provide 
 insights into the performance of scientific applications using OpenMP work
 sharing-loops versus OpenMP tasks. We investigate OpenMP task scheduling u
 sing the task constructs. We explore how to influence OpenMP task scheduli
 ng to mimic loop self‑scheduling behavior to achieve balanced execut
 ion and improve application performance. This study evaluates the effectiv
 eness of self-scheduling in load balancing task‑based applications.\
 n\nDomain: CS and Math
END:VEVENT
END:VCALENDAR