Toward Fault-Tolerant Adaptive Real-Time Distributed Systems

A monitoring approach to the problem of constructing fault-tolerant and adaptive real-time systems, based on the fail-signal processor, is described. Low error detection latency time is a primary goal. A fail-signal processor comprises an application processor along with a simple monitoring processor that detects abnormal functional or timing behaviour in the application processor; on such a failure the monitor issues a failure signal to other fail-signal processors and resets the application processor. The serviceow graph, used to specify real-time services, shows how a service is decomposed, redundantly designed, and structured to meet time-bounds. Information obtained from serviceow graphs along with run-time information provided by the fail-signal processors permits: (1) forward error recovery from failures in application processors; (2) avoidance or prediction of service timing failures; and (3) recon guration with graceful degradation. Avoidance of timing failures is based on adaptive scheduling.