The four measures I use to tell just how good my software systems perform are: reliability, success with unexpected stimuli, true system capacity, and the number of service calls. They all measure the system from the viewpoint of its field execution. The study of the dynamic behavior of software addresses these issues. Lawrence Bernstein is a technologist with Price Waterhouse and Chief Technologist for CyberExpress, a company bringing the Internet to China. He is a fellow of the ACM, IEEE and Ball State University, and is a member of the Russian Information Academy. Lawrence Bernstein retired from Bell Labs as Executive Director May, 1996, after 35 years of service. He was responsible for software technology and systems used to operate telephone networks, and the projects he managed automate operations of telecommunications service providers worldwide. |
 |
obust software is difficult to develop because it has a weak theoretical
foundation. Most of the theory that does exist focuses on the static behavior
of the software: requirements, estimation, software design, encapsulation,
data flow design, decomposition, structural analysis, and code complexity,
and concentrates on analysis of the source listing. There is little theory
on its dynamic behavior how it performs in the field, i.e., under load.
Even after we find and fix a bug, how do we restore the software to a known
state, one where we have tested its operation? For most systems, this is
impossible except with lots of custom design that is itself error-prone.