Redefining the Interface between Control and Real-Time Scheduling

Professor Paulo Tabuada, Director
The Cyber-Physical Systems Laboratory

Existing technological systems are remarkably efficient and reliable despite the wide variety of conditions they operate in. A major contributor to this success is the field of control engineering, which provides the scientific principles and design tools to guarantee desired levels of performance under uncertainty. The principles of control, and the concept of feedback in particular, are routinely used across the whole spectrum of engineering applications. Examples range from the very small scale, such as nuclear magnetic resonance spectroscopy and quantum information processing to the very large scale, such as electrical power grids and the internet, while including more familiar systems such as aircrafts, automobiles and consumer electronics.

Prof. Tabuada's group is currently conducting research at the boundary between control and real-time scheduling. Nowadays engineering applications continue to require a high degree of autonomy and performance under uncertainty. However, they pose new challenges to control engineering as they combine distributed information processing, networking and the usage of severely resource constrained embedded microprocessors. To increase the functionality and reliability of these systems, one has to question standing assumptions and develop new methods to divide processor time among competing tasks. The co-design of control and real-time scheduling was initiated less than a decade ago when researchers in the US and Europe started optimizing the frequency at which control tasks are executed in order to increase performance while guaranteeing a fair allocation of processor time. Prof. Tabuada's group is taking these ideas one step further by abandoning the paradigm of periodic execution of control tasks. "Consider a juggler tossing balls into the air", Tabuada says. "If he devotes an equal amount of attention to every ball, he will only be able to juggle a small number of them. However, if he pays more attention to the balls that are about to hit the ground, he will manage to juggle more balls while dropping none.

This is the same principle that we are applying to realtime scheduling. The decision of executing control tasks is based on the observed performance of the system being controlled rather than on an a-priori computed period. Strangely, this can be seen as an application of feedback to the software implementation of feedback control." The novel methods developed in Prof. Tabuada's group lead to a new scheduling paradigm where performance requirements are explicit while timing requirements are implicit and time varying according to operating conditions. This new paradigm, its mathematical principles and corresponding design tools are being developed at UCLA's Electrical Engineering Department under NSF support.