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:20181221T160726Z
LOCATION:D173
DTSTART;TZID=America/Chicago:20181111T112000
DTEND;TZID=America/Chicago:20181111T114500
UID:submissions.supercomputing.org_SC18_sess163_ws_works104@linklings.com
SUMMARY:LOS: Level Order Sampling for Task Graph Scheduling on Heterogeneo
 us Resources
DESCRIPTION:Workshop\nReproducibility, Scientific Computing, Scientific Wo
 rkflows, Workflows, Workshop Reg Pass, HPC, Data Intensive\n\nLOS: Level O
 rder Sampling for Task Graph Scheduling on Heterogeneous Resources\n\nWitt
 , Wheating, Leser\n\nList scheduling is an approach to task graph scheduli
 ng that has been shown to work well for scheduling tasks with data depende
 ncies on heterogeneous resources. Key to the performance of a list schedul
 ing heuristic is its method to prioritize the tasks, and various ranking s
 chemes have been proposed in the literature. We propose a method that comb
 ines multiple random rankings instead of a using a deterministic ranking s
 cheme.\n\nWe introduce L-Orders, which are a subset of all topological ord
 ers of a directed acyclic graph. L-Orders can be used to explore targeted 
 regions of the space of all topological orders.  Using the observation tha
 t the makespans in one such region are often approximately normal distribu
 ted, we estimate the expected time to solution improvement in certain regi
 ons of the search space. We combine targeted search and improvement time e
 stimations into a time budgeted search algorithm that balances exploration
  and exploitation of the search space.  In 40,500 experiments, our schedul
 es are 5% shorter on average and up to 40% shorter in extreme cases than s
 chedules produced by HEFT.
URL:https://sc18.supercomputing.org/presentation/?id=ws_works104&sess=sess
 163
END:VEVENT
END:VCALENDAR