PUMA

The PUMA project built on the work of Supernode 1 and continued in parallel with Supernode 2. It was again funded under European ESPRIT 2 (project P2701) and ran from 1989 to 1991. Whereas Supernode 2 aimed to develop scientific applications for the existing MIMD system developed under Supernode 1, the PUMA project aimed to enhance the underlying system architecture. The main goal of this project was to develop a switch network to allow communication between arbitrary pairs of processors, rather than the fixed topology of the previous generation. To exploit the new network, a new processor was designed based on the existing T8, and system software was developed to allow arbitrary point to point communications.

Our group at the University of Liverpool investigated how we could exploit the new network architecture in the design of distributed numerical algorithms.

My work in this project forms the basis of my PhD: The design of numerical algorithms for transputer systems.

Parallel Algorithms for the BFGS Update on a PUMA Machine

PUMA BFGS.
Tim Oliver. September 1991.

The Bitonic Sort on Transputer Architectures

PUMA sorting.
Tim Oliver. September 1991.

Occam on T8 and T9 machines: coding and conversion issues

PUMA conversions.
Gabriel Howard, Tim Oliver. August 1991.

A Communications Model for a PUMA Machine

PUMA communication modelling.
Tim Oliver. October 1990.

An Overview of Parallel Methods for Unconstrained Optimisation

PUMA optimization methods.
Tim Oliver. January 1990.

A Parallel Newton Method for Unconstrained Optimisation

PUMA Newton method optimization.
Tim Oliver. September 1989.

Writing numerical applications for the T9000

PUMA coding.
Rod Cook, Gabriel Howard, Tim Oliver, Bruce Stephens.

These documents are in Adobe Acrobat format.
You need to download the free Acrobat Reader to view and print them.