On the Utility of Supercomputers

Marc Snir, Computer Science, UIUC

Three years ago, DARPA started a major research program on “High Productivity Computing Systems”. The motivation for this program was a perception that supercomputer users have not benefited from the continued increase in raw computer performance to an extent that is commensurate with the performance increase. Rather than focusing on raw performance, DARPA would focus on increased supercomputer productivity. This immediately raises the question of measuring supercomputer productivity.

We present in this talk a framework based on Utility Theory that addresses this question. While this framework does not provide a direct answer, it indicates possible approaches to metric definition and indicates pitfalls to avoid.

Slides Audio