Inadequacy of Conventional Dynamic Recovery Mechanisms in the Presence of Temporary Failures

摘要

This paper shows that some implementations of fault-tolerant systems with dynamic error detection and reconfiguration mechanisms may not recover from certain types of temporary failures. An experiment is conducted to study the effect of temporary failures on the behavior of a dynamically redundant fault-tolerant system. The system is built out of LSTTL catalog parts. Transient failures are induced by reducing the power supply voltage; intermittent failures are induced loading nodes in the system. Reducing the power supply voltage produces common-mode failures that can be detected if the recovery mechanism produces high amplitude oscillations when its inputs are near the threshold level. Intermittent failures can be detected if the recovery mechanism detects errors before incorrect correct data is transmitted through the output devices. It is shown that the stuck-at fault model is inappropriate for the temporary failures injected into the system. Techniques are suggested that will guarantee detection of many transient and intermittent failures.