Adaptive System-Level Diagnosis in Real-Time

Abstract

Distributed real-time systems are subject to stricter fault-tolerance requirements than non-real time systems. This work presents an application of system-level diagnosis to a real-time distributed system as a first step in providing fault-tolerance. An existing algorithm for distributed system-level diagnosis, Adaptive_DSD, is converted to a real-time framework, establishing a deadline for the end-to-end diagnosis latency. Rate monotonic analysis is chosen as the framework for achieving real-time performance. The ADSD algorithm is converted into a set of independent periodic tasks running at each node, and a systematic procedure is used to assign priorities and deadlines to minimize the hard deadline of the diagnosis function. The resulting algorithm, Real-Time Adaptive Distributed System-Level Diagnosis (RT-ADSD), is fully compatible with a real-time environment, where both the processors and the network support fixed-priority scheduling. The RT-ADSD algorithm provides a useful first step in adding fault-tolerance to distributed real-time systems by quickly and reliably diagnosis node failures. The key results presented here include a framework for specifying real-time distributed algorithms and a scheduling model for analyzing them that accounts for many requirements of distributed systems, including network I/O, task jitter, and critical sections caused by shared resources.

This research is supported in party by the Office of Naval Research under Grant N00014-91-J-1304 and under a National Science Foundation Graduate Research Fellowship. Any opinions, findings, conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect the views of the Office of Naval Research or the National Science Foundation.