The quantitative approach, championed especially by Dave Patterson and John Hennessy through their widely-used textbooks, is now the overwhelmingly dominant methodology in computer architecture research. Practitioners of the method test architectural ideas by extracting numerical performance results from simulation models that run a standard set of benchmark programs. In this talk I will argue that while the quantitative approach has had a generally salutary effect on the field, it has become a kind of methodological straightjacket, encouraging formulaic papers and stifling qualitative inquiry. I will claim, moreover, that the methodology itself, unless applied with scrupulous care, can too easily lead investigators to unjustifiable conclusions. I will end by suggesting some qualitative directions for computer architecture, and by considering whether sheer performance might no longer be the right metric anyway.

Brief bio:

Doug Clark is a professor of computer science at Princeton. He received his Ph.D. in computer science from Carnegie-Mellon and his B.S. in engineering and applied science from Yale. Prior positions include a long period at the former Digital Equipment Corporation, a shorter one at Xerox PARC, and several summers as a harpsichord-maker's apprentice. Doug's principal research interests are in computer architecture and computer hardware.