Fuzzy belief-based supervision

Abstract

This thesis presents a new approach to automatic failure detection (supervision) of session-orientated, real-time software systems. The system being supervised is assumed to be specified in a formalism based on communicating extended finite state machines such as ITU-T SDL. The presented approach is a significant refinement of an existing belief-based supervision approach. Its novelty lies in the association of a feasibility factor with individual hypotheses about the state of the target software system. The feasibility changes over time according to the closeness of the hypothesis to the observed behaviour. The competition algorithms presented in the thesis decide which hypotheses are left and which are discarded. The approach allows for a continuous supervision, capable of resynchronization with the target system following occurrences of failures.

After a description of the approach, an experimental evaluation of the research results is presented. The target system in the evaluation was the control program of a small telephone exchange. Both the simulated exchange and the supervisor executed on a UNIX workstation.