Architecting Holistic Fault Tolerance

The efficiency and maintainability of fault tolerance mechanisms in a computer system has typically not been a major topic of concern, mostly because fault tolerance is a non-functional system requirement. This paper proposes a Holistic Fault Tolerance architecture, based on a centralised fault tolerance management, with related functionality distributed across the entire system. The most suitable error detection and error recovery strategies for a given application are chosen by a special crosscutting controller depending on error rates, system performance and resource utilisation requirements. We discuss the motivation for introducing this holistic fault tolerance architecture and reason about its benefits from the point of view of optimal system operation and improved maintainability. The advantages and possible implementation challenges of the proposed approach are demonstrated by a real-world application.