By Topic

Fussli: A portable framework for exploiting hybrid task, data and pipeline parallelism on multi-cores

Sign In

Cookies must be enabled to login.After enabling cookies , please use refresh or reload or ctrl+f5 on the browser for the login options.

Formats Non-Member Member
$31 $13
Learn how you can qualify for the best price for this item!
Become an IEEE Member or Subscribe to
IEEE Xplore for exclusive pricing!
close button

puzzle piece

IEEE membership options for an individual and IEEE Xplore subscriptions for an organization offer the most affordable access to essential journal articles, conference papers, standards, eBooks, and eLearning courses.

Learn more about:

IEEE membership

IEEE Xplore subscriptions

2 Author(s)
Xiaoye Wang ; Sch. of Inf. Sci. & Technol., Zhanjiang Normal Univ., Zhanjiang, China ; Ting Zhang

Parallelism is the most important mean to exploit the computation potential of multi-core processors. Real applications, particularly, commercial applications often have strong dependence that has to be respected. In order to achieve reasonably good performance, hybrid parallelism schemes usually need to be applied in these applications. Furthermore, parallel applications with task and pipeline parallelism are often developed in a target-orientated manner. Due to a parallel application is likely to run on different hardware platforms, a portable hybrid parallelism approach is attractive. This paper presents Fussli, a portable framework that allows the programmer to express hybrid parallelism with architecture-independent APIs and the framework the back-end takes advantage of hardware features. Along with parallelism expression provided by Fussli APIs, the profiling-driven parallelism mapper finds optimal parallelism configurations for different platforms. Experimental results on four typical applications and on two different multi-core platforms (an 8-core Intel Xeon a 9-core IBM Cell platforms) demonstrate the effectiveness of Fussli. We are able to obtain comparative or better performance for these applications compared to a recent proposed profiling-driven approach.

Published in:

Computer Application and System Modeling (ICCASM), 2010 International Conference on  (Volume:11 )

Date of Conference:

22-24 Oct. 2010