Database Reference
In-Depth Information
−
λ
t
e
Figure 6.5 Original curve and tangent line of
the International Disk Drive Equipment and Materials Association (IDEMA) stan-
dard (i.e., 4.38%/year
or
0.5%/1000 hours)
[38]
and the disk nominal life span of 5
years
[3]
, the range of
λ
t
is (0, 0.219). In this range, it can be
se
en that the deviation
of the tangent line is relatively small. With the decrease of
λ
t
, the deviation gets
even small
er
.
A
fter this substitution, equation
(6.1)
is further simplified into function
λλ
RR
(1)1
12
. Compared to equation
(6.3)
, the simplification of the complexity of
the equation is obvious. In addition to reducing the complexity of the equation, there
is another advantage of using the tangent line as a substitution. By solving equation
(6.1)
, the result (i.e., the checking interval values of the data file) is always conserva-
tively underestimated slightly, so that the deviation caused by the substitution does not
reduce the data reliability assurance that PRCR provides. In fact, by using the tangent
line substitution, the data reliability assurance PRCR provides is always higher than
the calculated result.
The execution time and accuracy rate of the algorithm for all four versions are
tested under the same data file and disk settings. Note that the accuracy rate stands
for the ratio between the results calculated by the optimized versions of the algorithm
and the original versions of the algorithm, which indicates the accuracy of the results
produced by the optimized versions of the algorithm. The results of the experiments
are shown in
Table 6.1
. The upper half of
Table 6.1
shows the average execution time
of all four versions of the algorithm. In addition, the number of checking interval
values calculated in each run of the algorithm is also shown in brackets for the ORV
and OPV versions of the algorithm, respectively. It can be seen that the optimized
versions of the algorithm outperform the original versions in several respects. First,
despite the same execution time incurred when data reliability is 99%, at which one
replica suffices to meet the data reliability and the codes of all versions of the algo-
rithm run are the same, the execution time of the optimized versions of the algorithm
is significantly smaller than that of the original versions of the algorithm. Second,
although the overall trend for all versions of the algorithm is that the execution time
increases with the increment of data reliability requirement, the execution time of the
optimized versions of the algorithm increases much slower than that of the original
versions. However, in the execution time part of
Table 6.1
, it also shows that, whether
=−
t
Search WWH ::
Custom Search