David Pearson

Research Interests

My thesis investigates highly scalable parallel computers consisting of very simple processors connected in a 3-dimensional mesh. The guiding vision of this work is of a time, perhaps 50 years hence, in which materials science has taken the place of computer architecture, and computers are crystals, with each processor a molecule in the lattice. Long before this goal can be realized, we can and should prepare for the ubiquitous parallelism it will offer. Our algorithms must heed the laws of physics and pay attention, as chip designers do now, to spatial layout and the (currently hidden) cost of communication. This can be accomplished by designing our algorithms for a 3-D mesh.

To pursue this vision requires both theoretical and practical work. So far, my work could be characterized as a feasibility study: I have produced a 3-D cellular architecture which could be efficiently realized with current hardware, a simulator for this architecture, algorithms, programs, and an operating system design for general-purpose computing. (I believe that general-purpose computers, not problems like protein structure, are the grand challenge of parallel architecture. Parallel computational power will not really succeed until it becomes a commodity and is sold for desktop machines or video games.)

Directions for future research include VLSI implementation of this architecture and the design of a programming language. Most widely-used languages hide the details of the machine's instruction set, but reflect the underlying Von Neumann architecture. I believe that this connection to the architecture has been a good thing for algorithm design, and to really exploit parallel machines we need a language for which the costs of operations are as easy to estimate as they are for C++ on a Von Neumann machine.

Publications


Computer Science Department
5133 Upson Hall
Cornell University
Ithaca, New York 14853-7501, USA

Email: pearson@cs.cornell.edu
Tel: (607) 255-9189
Fax: (607) 255-4428