Reliability Modeling of the MARS System:

A Case Study in the Use of Different Tools and

Techniques

Heinz Kantz

Institut f?ur Technische Informatik,

TU-Wien,

Vienna, Austria

Kishor Trivedi

Department of Electrical Engineering,

Duke University,

Durham, NC 27706, USA

accepted for publication at

4th Int. Workshop on Petri Nets

and Performance Models,

Dec. 1991, Melbourne, Australia

Abstract

Analytical reliability modeling is a promising method for predicting the reliability of different architectural variants and to perform trade-off studies at design time.

However, generating a computationally tractable analytic model implies in general an abstraction and idealization of the real system. Construction of such a tractable model is not an exact science, and as such, it depends on the modeler's intuition and experience. This freedom can be used in formulating the same problem by more than one approach. Such a N-version modeling approach increases the confidence in the results.

In this paper, we analyze the MARS architecture with the dependability evaluation tools SHARPE and SPNP, employing several different techniques including: Hierarchical modeling, stochastic Petri nets, folding of stochastic Petri nets, and state truncation. We critically examine these techniques for their practicability in modeling complex fault-tolerant computer architectures.