Physical Fault Models and Fault Tolerance - Models in Hardware Testing - page 245

Fig. 8.17 Distribution of exceptions provoked by the HI, PF and EI techniques

The results of the analysis conducted on the MARS self-checking node showed
that the four injection techniques are rather complementary, i.e., they exercise, to a
large extent, different types of error detection mechanisms. This advocates for the
application of various techniques to improve the confidence in the assessment of the
properties of a target fault-tolerant system.
Indeed, the extensive sets of test sequences significantly contributed to get con-
viction in the ability of the MARS nodes to sustain the “fail silence” property.
It was also shown that, beyond the hardware and system software error detection
mechanisms (EDMs), the application-level detection mechanisms are necessary for
achieving a very high coverage on the fail silence assumption. Indeed, although the
time-slice controller effectively prevents fail silence violations in the time domain,
fail silence violations in the value domain were observed for all four injection tech-
niques when double execution of tasks was not used.
We conclude by addressing some practical issues that have also to be taken into
account when selecting a fault injection technique. In addition to fault represen-
tativeness (i.e., the plausibility of the supported fault model with respect to actual
faults) that is one concern that is often raised in conjunction with fault injection
experiments, and for the study of which we provided a conceptual frame and objec-
tive insights, a wide range of criteria can be considered to assess the merits of the
fault injection techniques. Without any claim for an exhaustive analysis, we have
considered the following eight basic properties: reachability , controllability , with
respect to space and time , repeatability (with respect to experiments), reproducibil-
ity (with respect to results), non-intrusiveness , the possibility for time measurement
(e.g., error detection latency) and the efficacy to generate significant experiments.
A characterization of the considered fault injection techniques with respect to
these eight basic properties is shown in Table 8.6 . Further insights can be found
in Arlatetal. ( 2003 ) . For each property, the techniques are graded according to the
scale none , low , medium and high . It is worth noting that, although it is quite generic