Parallel Processing Architectures
Using PCs & Linux

Abstract

Commodity PC hardware offers surprisingly good performance at remarkably low cost; Linux is a very complete version of UNIX that is freely-available as source code and runs well on a variety of PC platforms. This tutorial covers a wide range of alternative approaches to using PC hardware and Linux to create high-performance parallel computers.

There are four fundamentally different approaches. The first approach is to implement SIMD within a register, as per the Intel MMX (MultiMedia eXtensions). A more traditional approach is to use SMP PCs as shared-memory MIMD systems. It is also possible to use a group of PCs connected by a network as a parallel processing cluster, and there are several alternative methods for making the network better support parallel processing with various different architectural models. Using a Linux system to "host" an attached parallel processor is the fourth technique.

In keeping with the theme of ISCA, emphasis will be placed on the different architectural models and implementations. The practical aspects of implementing and using these parallel systems will also be covered in detail.

Audience for Tutorial

The presentation targets teachers, researchers, engineers, and programmers who are interested in using low-cost PC hardware to provide a solid, UNIX-based, parallel computing platform. Both set-up of established configurations for "production" use and custom configurations to support research are covered. The tutorial presents theory, systems-level, and user-level views of how these systems are constructed and used.

Background for Tutorial

For maximum benefit, attendees should have basic familiarity with traditional parallel processing systems and issues. Detailed knowledge is not a prerequisite.

Biography

This tutorial will be given by Prof. Henry G. Dietz (Hank Dietz); an electronic version of his vita is available on-line at http://dynamo.ecn.purdue.edu/~hankd/EVITA

Outline

This tutorial is essentially an updated and slightly refocussed version of the full-day tutorial given August 16, 1996 at the International Conference on Parallel Processing. The complete set of slides, and support software, from that tutorial is available online at http://yara.ecn.purdue.edu/~pplinux/ICPP96/. For the ISCA tutorial, the changes include the addition of coverage of the new SWAR/MMX, a stronger focus on specific architectural configurations in general, and the removal of some implementation details (to fit the half-day format).

As for the 1996 ICPP tutorial, there will also be one or more floppies of support software, and we expect to make all materials available on the WWW sometime after the tutorial. Sites that we currently maintain and are relevant to this tutorial include:

• http://dynamo.ecn.purdue.edu/~hankd/SWAR/
• http://garage.ecn.purdue.edu/~papers/
• http://yara.ecn.purdue.edu/~pplinux/

This page was last modified May 21, 1997. [an error occurred while processing this directive]