Bounds on the Reliability of Fault-Tolerant Software Built by Forcing Diversity

Abstract

Fault tolerance via diversity has been advocated as a viable defence against common-mode failure in safety critical systems. The consequences of using diverse, redundant software components in fault-tolerant, software-based systems have been the subject of much research. In particular, Littlewood and Miller showed analytically how “forcing” diversity between redundant software components might achieve higher expected system reliability than if these components failed independently. But their theorems concerned very special scenarios. This paper examines various lower and upper bounds on the expected reliability of systems built by ”forcing diversity” and specify conditions for forced diversity to guarantee improved upper bounds on the system’s expected probability of failure on demand (pfd).