Browse Prior Art Database

Understanding and Exploiting Distribution

IP.com Disclosure Number: IPCOM000128279D
Original Publication Date: 1985-Dec-31
Included in the Prior Art Database: 2005-Sep-15
Document File: 21 page(s) / 67K

Publishing Venue

Software Patent Institute

Related People

Guy T. Almes: AUTHOR [+4]

Abstract

In most workstation based distributed systems, distribution is limited to client-server distribution: server processes reside on dedicated server machines and clients access the servers over the network. In this research project we explore a more sophisticated form of distribution, whereby several machines cooperate on the solution of a single problem. We also investigate the design of two support tools for such forms of distribution, namely efficient interprocess communication and monitoring tools. In particu-lar, we address three specific issues: (1) The execution of compute-bound programs on a collection of workstations connected by a local area network. This research assesses the potential of using a local network based system as a loosely cou-pled multiprocessor. We propose a communications paradigm for this multiprocessor, and we explore the kinds of problems that are suitable for implementation in this environment. (2) The efficiency of different forms of interprocess communication for distributed systems. We build different forms of interprocess communication, write a limited set of applications and compare their performance. We also experiment with different implementation strategies. (3) The design and evaluation of measurement and monitoring tools for distributed systems. This set of tools helps us both in understanding the effects of modifications to the interprocess communication mechanisms as well as in debugging and optimizing multi-machine programs. -2-

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 5% of the total text.

Page 1 of 21

THIS DOCUMENT IS AN APPROXIMATE REPRESENTATION OF THE ORIGINAL.

Understanding and Exploiting Distribution

Guy T. Almes Willy Zwaenepoel

Rice COW TR85-12

February 1985

Department of Computer Science Rice University P.O. Box 1892 Houston, TX 77251-1892

(713) 527-8101

Table of Contents

Project Summary 2 Project Description 3 1. Introduction 3 2. Distributed Program Execution 5 3. Efficiency of Different Interprocess Communication Mechanisms 12 4. Monitoring a Distributed System 17 5. Equipment and Personnel 19 Bibliography 21

Project Summary

In most workstation based distributed systems, distribution is limited to client-server distribution: server processes reside on dedicated server machines and clients access the servers over the network. In this research project we explore a more sophisticated form of distribution, whereby several machines cooperate on the solution of a single problem. We also investigate the design of two support tools for such forms of distribution, namely efficient interprocess communication and monitoring tools. In particu-lar, we address three specific issues:

(1) The execution of compute-bound programs on a collection of workstations connected by a local area network. This research assesses the potential of using a local network based system as a loosely cou-pled multiprocessor. We propose a communications paradigm for this multiprocessor, and we explore the kinds of problems that are suitable for implementation in this environment.

(2) The efficiency of different forms of interprocess communication for distributed systems. We build different forms of interprocess communication, write a limited set of applications and compare their performance. We also experiment with different implementation strategies.

(3) The design and evaluation of measurement and monitoring tools for distributed systems. This set of tools helps us both in understanding the effects of modifications to the interprocess communication mechanisms as well as in debugging and optimizing multi-machine programs.

-2-

1. Introduction

Rice University Page 1 Dec 31, 1985

Page 2 of 21

Understanding and Exploiting Distribution

We propose to investigate the parallel execution of programs on machines connected by a local area net-work and the design of support tools for such multi-machine computations. In particular, we address three topics:

(1) The parallel execution of compute-bound programs on a collection of workstations connected by a local area network.

(2) The efficiency of different forms of interprocess communication for distributed systems.

(3) The design and evaluation of measurement and monitoring tools for distributed systems.

These three topics are bundled together into a single proposal because they are all three aimed at taking advantage of the potential for distribution on a local area network and because together they can make efficient use of the equipment we propose to buy as part of this grant.

We will base our work on our previo...