Error latency measurements in symbolic architectures

A Hybrid Monitor Assisted Fault Injection Environment

“…In this example, faults were detected after an average of about 100 seconds, but half were detected in the first 35 seconds. This distribution is very similar to the distribution of latency associated with fault overwrites (figure 8b), and is characteristic of distributions found in past latency studies [Youn91]. The similarity in distributions implies that the probability that a local memory location will be read before it will be written does not significantly change with increasing dormancy.…”

“…Because this study used periodic sampling, the discovery times of only per manent faults could be measured. In [Youn91], a hybrid monitor approach to measuring error latency was applied to a TI Explorer II Lisp machine. The method is based on simulation of the error discovery process taken from a continuous trace of software-selected locations.…”