Direct Deposit -- When Message Passing Meets Shared Memory

Abstract

Assuming that most code for parallel and distributed systems will be written by some parallelizing compiler or a high-level parallel program generator tool chain the actual machine model (coherent shared memory or message passing distributed memory) will become less significant to the application writer. The focus will shift to providing the best possible data transfer services with the least complex and least expensive hardware technology. This thesis deals with the search for the best possible transfer services for parallel program generators and comprises several models to establish a better conceptional structure in the problem space of possible solution. As an experimental study, it also describes several experimental communications system on five different parallel or distributed computing platforms that serve as data points to verify the proper handling of alternatives and tradeoffs. In this revision of a Doctoral thesis, the author concedes that High Performance Fortran (HPF) and the well-promised ubiquitous programming generators that were to make parallel programming as easy as scripting a few GUI gadgets into an Internet browser did not really materialize. Instead, a large number of programmers coming from other disciplines have been trained and hired to program and maintain high-performance computing codes, like message passing parallel programs. For the validity and significance of the arguments in this thesis this trend is no problem. Ultimately, the programming of parallel and distributed applications by humans or programming through tools like program generators will raise exactly the same problems. The winner for the most efficient (i.e., the best data transfer services with the least amount of hardware support) is neither a pure coherent shared memory architecture nor a pure, coarse grain message passing distributed memory architecture. Looking at end-to-end transfers, the optimum lies in between the two extremes.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
May 19, 2000
Accession Number
ADA457623

Entities

People

  • Thomas M. Stricker

Organizations

  • Carnegie Mellon University

Tags

Communities of Interest

  • Cyber
  • Energy and Power Technologies

DTIC Thesaurus Topics

  • Application Software
  • Communication Channels
  • Communication Systems
  • Computer Programming
  • Computer Science
  • Computers
  • Data Transmission
  • Distributed Computing
  • Fish
  • Instruction Set Architecture
  • Multiple Access
  • Network Architecture
  • Operating Systems
  • Parallel Computing
  • Software Design
  • Three Dimensional
  • Two Dimensional

Fields of Study

  • Computer science

Readers

  • Database Systems and Applications
  • Parallel and Distributed Computing.
  • Systems Analysis and Design

Technology Areas

  • Space