Strategy switching Enhancing reliability in resource-constrained systems

Modern computer hardware may seem perfectly reliable. But, when radiation is involved (such as in space), or we have extreme sensitivity to unreliability (such as in safety-critical systems), we need to rethink that assumption. Any level of radiation can cause bit flips known as Single Event Upsets (SEUs), which are transient faults that corrupt data structures and drive the system into undefined behavior. Such behavior compromises the correctness of the entire system with potentially disastrous consequences.
But we can catch a breath: hardware faults can be dealt with. Through premium, radiation-tolerant hardware or through redundant execution, application code can be completely isolated from their impact. Radiation-tolerant hardware is effective, but also costly. And redundant execution may delay obtaining a result, which is unacceptable for systems that require predictable responses. And the final nail in the coffin: the added overhead of these mitigations drives up energy consumption, thus making particular use-cases impossible.
We can do better: in this thesis, we approach the competing constraints of reliability through a risk-based lens. By quantifying the acceptable amount of risk to the application, we can instead compromise between reliability, energy, timeliness and resource usage. Fault-tolerance can be dynamically adjusted to optimally leverage what little resources are available. We propose Strategy Switching, mechanisms designed to selectively apply redundancy in response to transient fault occurrences. We contribute a robust theoretical and practical framework for fault-tolerance, enhancing the reliability of resource-constrained real-time systems and setting the stage for further advances in resilient and adaptive computing.