On the Design and Analysis of Real-Time Computers.
C.M. Krishna · Deep Blue (University of Michigan) · 1984
The use of distributed computers to control life-critical systems, such as aircraft or nuclear reactors, has begun to be recognized as of enormous practical importance, and as a source of significant and difficult research problems. These problems include deriving objective, application-sensitive, and provably correct characterizations of computer performance, ensuring adequate fault-tolerance, maintaining processor synchrony, and allocating and scheduling tasks to processors so that rigid timing deadlines on task execution are met even in the face of up to a given maximum number of processor failures. Some recent results pertaining to these problems are presented. Performance measures appropriate to real-time computer-controllers are introduced, together with examples of their derivation. A tradeoff between redundancy and processing power in the context of real-time applications serves to provide some insight into these measures. Then follows a discussion of synchronization in real time, part of which is two theorems allowing the indefinite expansion of phase-locked clocks. After this, a discussion of the relative merits of reconfigurable and non-reconfigurable real-time systems and techniques is presented. The final problem addressed is the fault-tolerant allocation and scheduling of tasks to processors in a real-time multiprocessor. These applications, while interesting in their own right, also show the usefulness of the performance measures introduced earlier.