On finding an optimal combination of error detection mechanisms based on results of fault injection experiments

The paper analyzes the prerequisites for the determination of most efficient combinations of error detection mechanisms based on the results of fault injection experiments. Two crucial requirements are identified: (1) for the classification of the experiment outcome it is essential to observe whether a system failure actually occurs. Error detections should be recorded, but their interference with the flow of the experiment must be avoided. (2) New measures are introduced that allow formation and performance prediction of arbitrary subsets of mechanisms a posteriori, based on raw experimental data. To demonstrate the applicability of the approach, results of selected experiments are presented for a computer-board that features a variety of error detection mechanisms not common to off-the-shelf computers.