Reliability analysis of communicating recovery blocks

This paper studies two models for communicating recovery blocks (RB) with statistically-independently failing software versions where versions are executed sequentially. Model one considers two RB: RB-1 & RB-2, where RB-2 receives some data from RB-1. Thus, if a version in RB-2 fails then RB-1 has to rollback to its initial state. Model two considers two RB in conversation: both blocks must satisfy their respective acceptance tests before any of the blocks are allowed to exit from the conversation. Simple expressions for the reliability of the system are derived for models 1 and 2, and it is proved for them that the reliability of a RB consisting of versions ordered from smallest to largest based on failure probability is as reliable as any other list of the versions. Optimization models are developed for models 1 and 2. The paper presents efficient branch and bound procedures to solve the optimization models.