Home

About

Advanced Search

Browse by Discipline

Scientific Societies

E-print Alerts

Add E-prints

E-print Network
FAQHELPSITE MAPCONTACT US


  Advanced Search  

 
Efficient Primitives for Creating and Scheduling Parallel Computations
 

Summary: Efficient Primitives for Creating and
Scheduling Parallel Computations
Umut A. Acar Arthur Chargu´eraud Mike Rainey
Max-Planck Institute for Software Systems
{umut,charguer,mrainey}@mpi-sws.org
Abstract
We give a brief overview of our ongoing work on developing
efficient and expressive abstractions for programming mul-
ticore machines. We propose a programming interface for
expressing a parallel computation dynamically, as a directed
acyclic graph (DAG). The DAG consists of tasks and depen-
dencies between them. Because our interface lets the DAG
take shape as the computation unfolds, the programmer can
describe a variety of computations, including those express-
ible with existing parallel-computing paradigms, such as
fork-join, spawn-sync, and parallel futures. In some parallel
applications, such as parallel, load-balancing garbage col-
lectors and graph-connectivity algorithms, performance can
be improved by reducing the cost of synchronizing paral-
lel tasks. Our interface gives the programmer a few criti-

  

Source: Acar, Umut - Programming Languages and Systems Group,Max-Planck-Institut für Softwaresysteme

 

Collections: Computer Technologies and Information Sciences