Discriminating fault rate and persistency to improve fault treatment

Abstract

In this paper the consolidate identification of faults, distinguished as transient or permanerdintermittent, is approached. Transient faults discrimination has long been performed in commercial systems: threshold-based techniques have been practiced for several years for this purpose. The present work aims to contribute to the usefulness of the count-and-threshold scheme, through the analysis of its behaviour and the exploration of its effects on the system. To this goal, the scheme is mechanized as a device named acount, endowed with a few controllable parameters. a-count tries to balance between two conflicting requirements: to keep in the system those components that have experienced just transient faults; and to remove quickly those affected by permanent or intermittent faults. Analytical models are derived, allowing detailed study of a-count's behaviour; the actual evaluation, in a range of configurations, is performed by standard tools, in terms of the delay in spotting faulty components and the probability of improperly blaming correct ones.

Evaluation of Fault-Tolerant Multiprocessor Systems for High Assurance Applications

“…In the general case, as discussed in [4,11], increasing values of T decrease the number of wrongly removed healthy processors (as shown by Table 5), while the probability of erroneous computation tends to increase because faulty processors stay longer on-line. In the present setting, where wrong outputs (even if detected) cause costly mission failures, the second effect dominates.…”

“…The behaviour of -count has been analysed in [4,9,11] as a stand-alone mechanism. Two figures of merit have been evaluated: i) the time, D, between a (permanent or intermittent) fault occurrence in a processor u i , and its recognition by the threshold crossing of the pertinent count i ; ii) the wasted time, NU, spent by a processor idled after being wrongly signalled as faulty (normalized to the expected processor's life).…”

“…Our multiprocessor system is supposed to run tasks having the same duration, T t , which has been set to 600 sec. From the analysis of the -count mechanism's behaviour performed in [4,11], where the relation between the activation rate of intermittent faults and the decay ratio of the filtering function (K) has been studied, a significant pair of values has been The values just shown for the simulator settings are not intended to refer to any particular system; rather, some parameters, e.g. the fault rates, are representative of average values exhibited by a broad range of real systems; while others, as the mission times and confidence level, have been chosen to trade off precision and significance of simulated results vs. computational resources.…”

“…The -count mechanism, used as DM, is a simple count-and-threshold scheme presented in [4,11], which was designed to dis-criminate intermittent and permanent faults against low rate, low persistency transient faults.…”

Optimal discrimination between transient and permanent faults

“…Heuristics are suggested by intuitive reasoning, and then validated by experiment or modelling (see e.g. [1] for an assessment of a heuristic via modelling). Most on-line techniques [1,5,8,12,13] use thresholding schemes.…”