Towards comparability in evaluating the fault tolerance of safety critical embedded software

This thesis deals with the problem of obtaining meaningful and comparable dependability measures of software through the method of fault-injection. The thesis is specifically dedicated to safety-critical embedded software and its dependability property ‘fault-tolerance’ with respect to random faults affecting the machine instruction execution. For enabling comparability of the faulttolerance of different software on different hardware, a fault-injection method is derived and presented. The method is based on the idea of the so-called FARM sets which was originally presented for the characterization of physical fault-injection experiments. The collection of sets is broadened and adjusted to the herein considered object of evaluation ‘software in execution’. The software is thereby conceived as a process, and the fault set F is devised accordingly by means of a universal microprocessor model. The resulting set is software-overlapping and forms a mutual basis regarding the fault input among the software. In conjunction with the other sets presented, a fault-injection method allowing for – as far as possible – comparable fault-tolerance measures is constructed. Therewith is created a methodical fundament for fault-injection experiments that aim at evaluating the fault-tolerance of safety-critical embedded software affected by hardware faults.