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:20181221T160904Z
LOCATION:C2/3/4 Ballroom
DTSTART;TZID=America/Chicago:20181115T083000
DTEND;TZID=America/Chicago:20181115T170000
UID:submissions.supercomputing.org_SC18_sess324_post224@linklings.com
SUMMARY:Enabling High-Level Graph Processing via Dynamic Tasking
DESCRIPTION:Poster\nTech Program Reg Pass, Exhibits Reg Pass\n\nEnabling H
 igh-Level Graph Processing via Dynamic Tasking\n\nDrocco, Castellana, Minu
 toli, Tumeo, Feo\n\nData-intensive computing yields irregular and unbalanc
 ed workloads, in particular on large-scale problems running on distributed
  systems. Task-based runtime systems are commonly exploited to implement h
 igher-level data-centric programming models, promoting multithreading and 
 asynchronous coordination for performance. However, coping with dynamic wo
 rkloads (e.g., those yielded by large-scale graph processing) is challengi
 ng.\n\nIn this work, we took an exploratory approach to overcome some typi
 cal bottlenecks in tasking systems. In particular, we propose 1. a novel t
 ask allocator based on dynamic per-thread allocation and all-to-all recycl
 ing networks, and 2. a reservation-free remote spawning schema, based on r
 eceiver-side buffering and back-pressure feedback/sensing to avoid overflo
 ws.\n\nAs a proof of concept, we implemented the proposed techniques under
 neath a high-level library of distributed C++ containers. Preliminary expe
 rimental evaluation shows consistent scalability, a neat improvement in pe
 rformance (e.g., 1.5x speedup with respect to the original code over an 8M
 -nodes graph), and less sensitiveness to parameter tuning.
URL:https://sc18.supercomputing.org/presentation/?id=post224&sess=sess324
END:VEVENT
END:VCALENDAR

