Deriving Dependability Measures of Measurements Recorded in a Matrix

Dependability benchmarking is meant to measure system characteristics like availability, reliablity, data integrity etc. Todays systems are working at high levels of these characteristics. Evaluation of these characteristics demands to inject faults forcing fault tolerant mechanisms to exercise their tasks. Observing the response of the system leads to measurements assessing the quality of these mechanisms. Our paper’s focus is not on how to create a special dependability benchmark but on how to deduce significant dependability characteristics out of fault injection based measurements. We disclose which information we need for a general dependability benchmark, from whom they should be supplied and, finally, how to derive assessments of dependability metrics from this information. Our method is universally applicable to all fault injection based dependability benchmarking methods. Using one method for the presentation of dependability has several advantages: E.g. benchmarks become comparable to each other, benchmarkers faster learn how to interpret similarly looking results.