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:20210916T132456Z
LOCATION:
DTSTART;TZID=Europe/Stockholm:20210706T173000
DTEND;TZID=Europe/Stockholm:20210706T190000
UID:submissions.pasc-conference.org_PASC21_sess182_post155@linklings.com
SUMMARY:P29 - Don’t Compete, Let’s Cooperate: A Cooperati
 ve Scheduling Approach
DESCRIPTION:Poster\n\nP29 - Don’t Compete, Let’s Cooperate: A&
 nbsp;Cooperative Scheduling Approach\n\nEleliemy, Ciorba\n\nScientifi
 c application developers are concerned with improving their
  applications' execution time and less concerned with increasing 
 ;HPC systems' utilization. HPC system operators priori
 tize increased system utilization over improved individual applicatio
 n performance. These goals yield an ecosystem where each applica
 tion runs on a set of exclusively-allocated resources for the al
 lotted time (space policy) and applications are assumed to efficientl
 y utilize the allocated resources (balanced execution).  In
  practice, resources tend to be idle during application execution due
  to load imbalance or pre- and post-processing. Moreover, application
 s may also exhibit variable computational needs during execution. In 
 this work, we propose an approach to enable cooperation&nbs
 p;between, currently independent, batch and application schedule
 rs. Our cooperative scheduling approach enables applic
 ation schedulers to share their allocated but idle computing resources thr
 ough the batch system.  With resource sharing being driven 
 by the application schedulers, this approach avoids the occurren
 ce of resource shrinking operations and the associated performance penalti
 es that are typical of dynamic resource and job management syste
 ms. Realistic batch (Slurm-based simulator) and application (SimGrid-
 based simulator) schedulers  were used to implement the proposed appr
 oach. Results of simulating HPC workloads from the public workload archive
  showed that our approach increases system utilization by 10% without affe
 cting applications' performance.
END:VEVENT
END:VCALENDAR