Metacomputing on Commodity Computers

Abstract

The advantages of utilizing networks of commodity computers as a platform to execute compute-intensive parallel programs are well known: commodity computers are relatively cheap, widely available, and mostly underutilized. For example, here at the Department of Computer Science at New York University, students have access (and can login) to over 200 workstations. At the same time, I have seen individuals use a single workstation to run programs that take hours to complete. These programs could have been parallelized and made to execute on the network of available workstations which would have completed in just a few minutes. But programs were not parallelized|this dissertation addresses the reason behind this. Utilizing networks of commodity computers to execute parallel programs is not an original idea|much research has been devoted to this topic, and many software tools have been built for developing such programs. Given that there are relatively small number of widely available parallel programs that run on networks of workstations, a valid question to ask is: if the hardware is widely available (which it is) and if there are tools for building parallel programs (which there are), then why aren't most programs able to run on networks of workstations? I claim that major contributing factors are the complexity involved in software development and the extra effort needed to execute such programs. That is, existing software tools make the development and execution of distributed parallel programs possible but not always feasible; as a result, the added complexity outweighs the gains. This dissertation presents a set of techniques for making parallel programs easy to design, build, and execute on networks of commodity computers. Furthermore, it presents a series of software systems to validate the feasibility of these techniques.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
May 01, 1999
Accession Number
ADA598393

Entities

People

  • Arash Baratloo

Organizations

  • New York University

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Computations
  • Computer Program Documentation
  • Computer Program Reliability
  • Computer Programming
  • Computer Programs
  • Computer Science
  • Computers
  • Graphical User Interface
  • Local Area Networks
  • Network Protocols
  • Networks
  • Operating Systems
  • Parallel Computing
  • Programming Languages
  • Shell Scripts
  • User Interface
  • Web Browsers

Fields of Study

  • Computer science

Readers

  • Computer Science.
  • Database Systems and Applications
  • Distributed Systems and Data Platform Development