Dissertation/Thesis Abstract

Tideflow: A dataflow-inspired execution model for high performance computing programs
by Orozco, Daniel A., Ph.D., University of Delaware, 2012, 163; 3527015
Abstract (Summary)

Traditional programming, execution and optimization techniques have been shown to be inadequate to exploit the features of computer processors with many cores.

In particular, previous research shows that traditional paradigms are insufficient to harness the opportunities of manycore processors: (1) traditional execution models do not provide constructs rich enough to express parallel programs, (2) traditional analysis tools allow little insight into the performance of parallel programs and (3) traditional programming and execution tools only offer awkward ways to execute parallel programs.

This thesis addresses those problems with the introduction of TIDeFlow, a parallel execution model aimed at efficient execution and development of High Performance Computing (HPC) programs in manycore processors.

The following are the main contributions of this thesis: 1. The formulation of a parallel execution model that is able to exploit the features present in manycore processors. 2. The development of several highly scalable algorithms and a technique to analyze their throughput. 3. The implementation of the TIDeFlow toolchain, including a programming model and a distributed runtime system.

Indexing (document details)
Advisor: Gao, Guang R.
Commitee: Gao, Guang R., Li, Xiaoming, Taufer, Michela, Yang, Chengmo
School: University of Delaware
Department: Electrical and Computer Engineering
School Location: United States -- Delaware
Source: DAI-B 74/02(E), Dissertation Abstracts International
Source Type: DISSERTATION
Subjects: Computer Engineering
Keywords: Dataflow, Distributed runtime systems, Execution models, Fine grain execution, Parallel computing, Parallel programming
Publication Number: 3527015
ISBN: 978-1-267-61132-1
Copyright © 2019 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest