home · mobile · calendar · defenses · 2003-2004 · 

Thesis Defense - Barthelmess

ThreadMill: A Highly Configurable Architecture for Human Communication Analysis Applications
Paulo Barthelmess
Computer Science PhD Candidate

Computer applications that analyze human communication are shaped by the intrinsic complexity of the task and evolved to employ techniques that at an abstract level have similarities, independently of the specific modality that is targeted.

Closer examination of applications in this domain reveals opportunities for enhanced performance through concurrent execution.

While opportunities for concurrent execution are plenty, and hold promise of enhanced performance and quality of results, developing software that takes advantage of it can be challenging. High-performance languages and libraries supporting parallelism and distribution are concerned mostly with efficiency of execution, and typically do not consider system-level abstractions.

While this class of solutions permits a fine level of control over concurrency aspects, communication, coordination and concurrency concerns become interwoven with application code. This makes it hard to reuse, evolve and efficiently re-deploy applications, particularly on execution environments that differ significantly from the one that an application was originally designed for.

This thesis proposes a generic component software architecture - ThreadMill - that facilitates the development and deployment of efficient distributed and parallel applications that process large volumes of streamed data. ThreadMill provides a powerful set of abstractions that are conducive to efficient execution.

The chosen abstractions hide the complexity of distributed and parallel programming to facilitate development, but at the same time give developers fine control over concurrency aspects.

Committee: Clarence (Skip) Ellis, Professor (Chair)
Michael Burl, Assistant Professor
Bryan Pellom, Research Assistant Professor
Brenda Schick, Speech, Language & Hearing Sciences
Valerie Taylor, Texas A&M University
Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:20)