Recovery in multicomputers with finite error detection latency

In most research on checkpointing and recovery, it has been assumed that the processor halts immediately in response to any internal failure (fail-stop model). This paper presents a recovery scheme (independent checkpointing and message logging) for a multicomputer system consisting of processors having a non-zero error detection latency. Our scheme tolerates bounded error detection latencies, thus, achieving a higher fault coverage. The simulation results show that for typical detection latency values, the recovery overhead is almost independent of the detection latency.