Take a look at the new version of this page: [ beta version ]. Tell us what you think.

ABSTRACT

Being named recipient of the 2004 ACM Sigmetrics Achievement Award has done several things to me. It brought me surprise that I would be singled out from the many people who have made significant and sustained contributions to the field of performance evaluation. It also brought me deep appreciation for all the students and colleagues with whom I have worked and come to know as friends over the years. Finally, it has caused me to ponder and reminisce about many of the research projects and consulting studies in which I have participated.In this talk, I will describe various systems I have used and studied, various applications of interest, and various models that I, and others, have used to try to gain insights into the performance of systems. Some lessons of possible future relevance that emerge from this retrospective look at a wide variety of projects are the following: <ol>

Exact Answers Are Overrated -- While exact solutions of mathematical models are intellectually satisfying, they are often not needed in practice.

Analytic Models Have a Role -- Analytic models can be used to obtain quick and inexpensive answers to performance questions in many situations where neither simulation nor experimentation are feasible.

Assumptions Matter -- Subtle changes to the assumptions that underlie an analytic model can substantially alter the conclusions reached based on the model.

</olAfter considering all the methods of analysis, simulation and experimentation, my recommendation for the very best means to attain substantially improved computer system performance is: Wait thirty years!