College of Computing Sciences

Recoverability Preservation

We consider a system that is supposed to perform a function F to meet some requirements specification R (represented by a relation).   If, due to some fault in the system, it fails to perform function F but performs some function F', then we need to know how far off-course we have gotten, and what (if anything) can be done to deal with the fault.  We know that faults may fail to be sensitized at all; that even when they are sensitized they mail or may not propagate (and are masked instead); and that even when they are sensitized and propagated they may or may not violate the requirements specification R (because R may be vastly non deterministic).  We add another level to this hierarchy:  A fault may be sensitized, may propagate, may violate the requirements specification R, yet still produce a recoverable state; in those cases, it is possible to still produce a correct outcome by applying a recovery routine.  We have mathematical results that characterize a function F' that is guaranteed to produce recoverable states, even when it is not guaranteed to produce correct states, nor maskable states, nor specification-admissible states.  In this project we are interested to investigate implications and applications of these results.  A summary of these results is given in this paper.