Minimum replication for meeting the data reliability requirement - Reliability Assurance of Big Data in the Cloud - page 43

From the aspect of execution time, the results show that the time for determining
the minimum replica number (i.e., one or two replicas) is less than 1 millisecond. The
time for predicting the longest storage duration of the data file for meeting the data reli-
ability requirement is at the magnitude of several milliseconds to tens of milliseconds,
which varies according to the data reliability requirement and whether the optimized
formulas are applied. Specifically, when the optimized formulas are applied in the
algorithm, the execution time could be significantly reduced. In general, the results
indicate that the minimum replication calculation approach is able to effectively deter-
mine the minimum number of replicas for the Cloud data storage as well as predicting
the longest storage duration of the data file for meeting the data reliability requirement
while incurring a relatively small computing overhead in terms of execution time.
From the aspect of accuracy rate of the optimized algorithm output, among differ-
ent versions of the algorithm, the results show that the output of the optimized algo-
rithm is very close to the original algorithm. The accuracy rate is lower (90%) when
the data reliability requirement is low, and quickly increases with the increment of the
data reliability requirement (99.9% when the data reliability requirement is 99.999%
per year). In general, this result indicates that our optimization to the minimum rep-
lication calculation approach could generate results that are of little difference com-
pared to the original one.
5.4 Summary
Based on the generic data reliability model presented in Chapter 4 , in this chapter we
presented our approach of calculating the minimum replication for meeting a given
data reliability requirement. We first presented the formulas for determining the min-
imum replica number and also presented our optimization solution for solving the
equations so that the prediction process of the longest storage duration of the data
file can be conducted efficiently to reduce the overhead for managing the Cloud data.
Then, we discussed the issue of using the minimum replication as a benchmark for
evaluating the cost-effectiveness and data reliability assurance ability of a replication-
based storage system. Finally, we briefly presented the satisfactory outcomes of the
evaluation for the minimum replication calculation approach.