Empirical evaluation of accuracy of mathematical software used for availability assessment of fault-tolerant computer systems
Kharchenko, V. S.; Popov, P. T.; Odarushchenko, O.; Zhadan, V.(2012)
International Group on Reliability
Dependability assessment is typically based on complex probabilistic models. Markov and semi-Markov models are widely used to model dependability of complex hardware/software architectures. Solving such models, especially when they are stiff, is not trivial and is usually done using sophisticated mathematical software packages. We report a practical experience of comparing the accuracy of solutions stiff Markov models obtained using well known commercial and research software packages. The study is conducted on a contrived but realistic cases study of computer system with hardware redundancy and diverse software under the assumptions that the rate of failure of software may vary over time, a realistic assumption. We observe that the disagreement between the solutions obtained with the different packages may be very significant. We discuss these findings and directions for future research.
City Research Online (http://openaccess.city.ac.uk/1521/1/Popov_GnedenkoForum.pdf)