Electronic Theses and Dissertations

Date of Award

2017

Document Type

Dissertation

Degree Name

M.S. in Engineering Science

Department

Computer and Information Science

First Advisor

Byunghyun Jang

Second Advisor

Feng Wang

Third Advisor

Philip Rhodes

Relational Format

dissertation/thesis

Abstract

Although many general purpose workloads have been accelerated on graphical processing units (gpus) over the last decade, other applications whose runtime behaviors are dynamic and irregular such as ones based on trees and graphs have suffered from serious workload imbalance problem caused by architectural differences between cpu and gpu processors. In this thesis, we propose a work-stealing framework to overcome such problems. Our proposed framework allows cpu and gpu threads to steal tasks from each other as well as within the same device by leveraging fine-grained data sharing and thread communication feature available on modern cpu-gpu heterogeneous systems. The implementation of bfs application on the top of our framework achieves a minimum of 8.5% performance improvement over the one with coarse-grained task partitioning scheme. It also achieves 16% performance improvement on average over its non-stealing counterpart.

Concentration/Emphasis

Emphasis: Computer Science

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.