Analyses Using Stochastic Reward Nets
Lorrie A. Tomek, Kishor Shridharbhai Trivedi · 1995
In this chapter, we examine the power of stochastic reward nets (SRNs), a variant of stochastic Petri nets, to model fault tolerant software systems. The three types of software fault tolerance we examine are:N -version programming, recovery blocks, andN self-checking programming. SRNs allow each fault tolerance technique to be specied in a concise manner. Complex dependencies, such as common-mode versus separate failures and detected versus undetected failures, associated with these systems are incorporated into each SRN model without undue complication. Underlying an SRN is a Markov reward model. Each SRN is automatically converted into a Markov reward model from which steady-state, transient, cumulative transient, and sensitivity measures are easily obtained. We study several measures, including reliability, safety, and performance measures, which are of interest in the evaluation of fault tolerant software techniques. We parameterize our model to account for common-mode failures between variants using distributions from experimental data, and then provide numerical results using the stochastic Petri net package (SPNP) [Cia89].