Fault-Tolerant Dynamic Systems

Modular redundancy (system replication) and other traditional techniques for fault tolerance in dynamic systems are expensive, and rely heavily — particularly in the case of systems operating over extended time horizons — on the assumption that the error-correcting mechanism (e.g., voting) is fault-free. Herein, we construct redundant dynamic systems in a way that achieves tolerance to transient failures in the state transition and error-correcting mechanisms. Using unreliable system replicas and unreliable voters, we construct redundant dynamic systems that evolve with low probability of failure for any given (finite) time-interval. For the special case of linear finite state machines (LFSM’s), we combine these techniques with low-complexity errorcorrecting codes and efficiently protect identical unreliable LFSM’s that operate in parallel on distinct input sequences. The approach requires only a constant amount of redundant hardware per machine and achieves a probability of failure that remains below any given bound for any pre-specified (finite) time-interval.