Dynamic management of resources is essential to building high performance
systems. Often, these dynamics involve feedback loops that are difficult
to design and analyze. It turns out, however, that other engineering
disciplines (e.g., mechanical and electrical engineering) have good
approaches for dealing with analogous problems in their field. For the
most part, these approaches are based on control theory.
This tutorial provides an introduction to classical control theory (linear,
deterministic, time invariant) and presents two applications to computer
science. We begin with a review of continuous time, linear system theory.
Included here is (a) use of the Laplace transform to describe time domain
functions, (b) the final value theorem, (c) modeling with block diagrams,
(d) transient response analysis, and (e) conditions for stability and
oscillatory response. This material is presented in the context of a
single class fluid flow model. Considered next are the basic control
actions: proportional, integral, and differential control. The
characteristics of these controllers are analyzed in terms of linear system
theory to identify issues of stability, bias, and settling time. Root
locus analysis is discussed as well. We then address control in discrete
time and related issues (e.g., sampling times). We also consider system
identification (determining the model for the system being controlled). An
extensive example is presented of a controller for an email server. We
show that root locus analysis correctly predicts where controllerinduced
oscillations occur. The tutorial concludes by touching on a number of
topics of use in applyikng control theory to computer science. Included
here are: frequency response techniques, state space (multiple input,
multiple output) models, and nonlinear control.

Joseph L. Hellerstein <hellers@us.ibm.com>
Joseph L Hellerstein is a research staff member at the IBM Thomas J Watson
Research Center where he manages the systems management research
department. Dr. Hellerstein received his PhD from the University of
California in Los Angeles. Since then his research has addressed various
aspects of managing service levels, including: predictive detection,
automated diagnosis, expert systems, and the application of control theory
to resource management. Dr. Hellerstein has published approximately 50
papers and an AddisonWesley book on expert systems.
Sujay Parekh <sujay@us.ibm.com>
Sujay Parekh received his B.S. degree in Computer Science from Cornell
University in 1993, and his M.S. from University of Washington in 1996.
From 1993 to 1994, he was a member of the Technical Staff at Oracle
Corporation. He is currently a Research Associate at the IBM T.J. Watson
Research Center while also working on his PhD at University of Washington.
His research interests include adaptive algorithms, systems management and
artificial intelligence.
