Browse Prior Art Database

DATAFLOW COMPUTER ARCHITECTURE: RESEARCH AND GOALS

IP.com Disclosure Number: IPCOM000128324D
Original Publication Date: 1978-Dec-31
Included in the Prior Art Database: 2005-Sep-15
Document File: 22 page(s) / 67K

Publishing Venue

Software Patent Institute

Related People

Arvind: AUTHOR [+3]

Abstract

The primary objective of the proposed research is to con-tinue the definition and evaluation of an effective architecture for a general-purpose dataflow computer composed of large numbers (hundreds or perhaps even thousands) of small LSI processors. thereby taking new advantage of LSI technology. We also propose to study associated system problems with the eventual goal being a specification of a complete dataflow computer system.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 6% of the total text.

Page 1 of 22

THIS DOCUMENT IS AN APPROXIMATE REPRESENTATION OF THE ORIGINAL.

DATAFLOW COMPUTER ARCHITECTURE: RESEARCH AND GOALS

by

Arvind Kim P. Gostelow

Technical Report #113

Department of Information and Computer Science University of California, Irvine February 6, 1978

*This work is supported by NSF Grant UCI Dataflow Computer Architecture Project MCS76- 12460.

I. The Objective, Significance, and Research Methods to be Employed

1. Objective

The primary objective of the proposed research is to con-tinue the definition and evaluation of an effective architecture for a general-purpose dataflow computer composed of large numbers (hundreds or perhaps even thousands) of small LSI processors. thereby taking new advantage of LSI technology. We also propose to study associated system problems with the eventual goal being a specification of a complete dataflow computer system.

2. Basis and Significance of the Proposed Work

2.1 Basis

Many other proposals (e.g., the Holland machine, Illiac IV, Hydra, Hypercube) have been forwarded in an attempt to synthesize a single large machine from several small processors which cooperate in a concerted effort on a single computation. Almost all of these proposals (some exceptions are [Chamberlin-71, Dennis & Misunas"74, Rumbaugh-77, Sonnenburg & Irani-741) have failed to recognize that such a machine, only recently made possible by new advances in technology, must itself be founded upon new principles of computation. This is the fundamental premise upon which this proposal stands. We claim that the problems involved in utilizing the new technology are not related simply to providing a proper interconnection mechanism, or to designing a machine which, for example, can efficiently manipulate arrays. Rather, the problems are due to one of the fundamental Page 2

premises of computer design: the von Neumann model of computation and its basic principles of a (centralized) sequential control and direct programmer manipulation of memory cells. In

University of California, Irvine Page 1 Dec 31, 1978

Page 2 of 22

DATAFLOW COMPUTER ARCHITECTURE: RESEARCH AND GOALS

place of the von Neumann model, we have adopted the opposing principles of dataflow [Dennis- 73, Karp & Miller-66, Rodriquez-69, BAhrs-72]:

1. operations execute when and only when the operands required become available (asynchrony),

2. Computation is based on the values produced rather than on where those values are kept (functionality, i.e., the absence of side-effects). By adopting these principles, we can realize the asynchronous execution of programs without the need for parallel programming constructs (e.g., parbegin - parend) or program analysis of any kind. our approach has been to develop simultaneously a base machine language and a higher-level dataflow programming language called ID (for Irvine Dataflow) [Arvind, Gostelow, & Plouffe-76]. Operationally, programs are written only in ID. They are then compiled into the base language (Section II - Semantic...