Degree Type

Dissertation

Date of Award

2012

Degree Name

Doctor of Philosophy

Department

Computer Science

First Advisor

Carl K. Chang

Abstract

In recent years, as we come closer to approaching physical limits in making smaller (and faster) computer processors, focus has instead been turned toward including multiple processor cores in each device. While this technically allows for more computational power as compared with only one traditional processor core, conventional software can typically only make use of a single processor. Furthermore, we see an increasing number of stream programs that process streams of data such as a stream of images or audio. For stream programs to effectively utilize multi-core processors, multithreading is the key, but it may be difficult to implement in practice depending on the complexity of the programs.

We present SAPPHIRE: Semi-Automated Parallel Programming in Heterogeneous Intelligent Reconfigurable Environment, a middleware and SDK for developing multithreaded stream programs. In this middleware, we implement our semi-automated program construction technique which is designed to aid in writing multithreaded software by reducing needed complexity and lines of code written by software developers.

We also present a novel static task-scheduling algorithm for stream programs with heterogeneous implementation choices. Our algorithm is capable of scheduling stream programs with provably near-optimal results given a specific set of assumptions, without requiring the unrolling of the task graph. Unrolling the task graph greatly increases the size of the input to the NP-Complete part of the task-scheduling problem as in related work.

Finally, we present two case study programs implemented using SAPPHIRE. One case study, EM-Capture, has analyzed over 50 billion frames of endoscopy video in real-time in a real hospital, discerning over 71,000 unique endoscopy procedures. The other case study, EM-Feedback-RT, is a collaborative extension to EM-Capture, and is an attempt to provide real-time quality analysis feedback to physicians during a colonoscopy exam.

Copyright Owner

Sean Stanek

Language

en

Date Available

2015-06-01

File Format

application/pdf

File Size

184 pages

Share

COinS