The Effects of Workload Input Domain On Fault Injection Results

It is well known that workload has significant impact on dependability measures obtained from fault injection experiments [3][5]. The effects on the outcome of fault injection experiments when the input data to be processed by the workload is altered, should also be considered [2]. To investigate the effects of workload input domain on fault injection results (e.g. error coverage), an experimental set-up using the FIMBUL tool [4] was employed. The problem of accurately estimating error coverage, if the effects of input domain are significant, could be solved by performing several fault injection campaigns using different input sequences. Since this would be very time consuming, a methodology to speed up the process is presented. The methodology involves predicting error coverage for different input sequences based on fault injection experiments conducted using another input sequence.