This paper takes a novel approach to fault detection, advocating a “monitor” or shadowing process that can detect a certain class of defects in the monitored process. The errors detectable are those which cause a control-flow defect in the monitored machine. How this is done without 100 percent replication of the monitored process by the monitor is what’s interesting here: The instructions are classed into groups and an incorrect group transition is detected by having the monitoring process signal when certain specified transitions occur.
This may be the first application of the Hartmanis and Stearns partition pair algebra developed in [1]. And it might even be practical] From the results presented in the paper, it appears possible to detect (and, hence, immediately correct for) a nontrivial fault-set. It even turns out that the monitoring machine is smaller than the one monitored. This may signal an advance in methods for attaining ultra-reliable computers through a measure of self-diagnosis.