Path-based error coverage prediction

Previous studies have shown that error detection coverage and other dependability measures estimated by fault injection experiments are affected by the workload. The workload is determined by the program executed during the experiments, and the input sequence to the program. In this paper, we present a promising analytical post-injection prediction technique, called path-based error coverage prediction, which reduces the effort of estimating error coverage for different input sequences. It predicts the error coverage for one input sequence based on fault injection results obtained for another input sequence. Although the accuracy of the prediction is low, path based error coverage prediction manages to correctly rank the input sequences with, respect to error detection coverage, provided that the difference in the actually coverage is significant. This technique may, drastically decrease the number of fault injection experiments, and thereby the time, needed to find the input sequence with the worst-case error coverage among a set of input sequences