
540 Large Scale and Big Data
We take a combined approach instead. This method depends on the best-effort
analysis of the whole process of MapReduce processing framework, which will result
in the following cost function:
TS
DC
ii
i
k
(, ,) (, ,)D
=
∑
h
1
, (17.1)
where h
i
(S, D, C) are possibly some nonlinear transformations of the input factors S,
D, and C, which are the time complexity of sequential processing components in the
system, and β
i
are the component weights, different from application to application.
h
i
(S, D, C) are obtained through the analysis of the MapReduce processing compo-
nents, while β
i
will be learned for a specic application ...