Information Technology Reference
In-Depth Information
F
IG
. 11. Reliability growth comparison for different workload measures for SMU/SEAS.
where
λ
0
and
λ
T
are the initial and final failure rates, respectively, estimated
by a fitted SRGM. Complete elimination of all potential defects would result in
ρ =
1, and no defect fixing would result in
ρ =
0. Normal reliability growth
is associated with
ρ
values ranging between these two extremes, with larger
ρ
values associated with more reliability growth. (1
− ρ
) gives us the ratio be-
tween
λ
T
and
λ
0
, or the final failure rate as a percentage of the initial failure
rate.
We fitted the widely used Goel-Okumoto (GO) model
[17]
introduced in Section
2
to relate cumulative unique failures (
m(t )
) to cumulative workload measurements (
t
)
in the formula:
1
−
e
−bt
m(t ) = N
.
Table VI
summarizes these modeling results, giving estimated model parameters
N
and
b
,
λ
0
,
λ
T
, and
ρ
.The
ρ
values based on models using different workload mea-
surements indicate that potential reliability improvement ranges from 57.9 to 74.8%
in purification levels. In other words, effective web testing and defect fixing equiv-
