Big Data Processing Systems - Cloud Data Management - page 137

Fig. 9.3 Decision tree for choosing between various join strategies on the MapReduce framework

coordinator

driver

process

fork

communication

mapper
mapper
mapper

split

reducer DFS file/chunks

split
split

internal transfer

reducer

DFS input/output

remote transfer

split
split

reducer

mapper

merger

output
output
output

mapper

merger

split
split
split
split
split

reducer

mapper
mapper
mapper

merger

reducer

reducer

Fig. 9.4 An overview of the Map-Reduce-Merge framework

To tackle the limitation of the extra processing requirements for performing join
operations in the MapReduce framework, the Map-Reduce-Merge model [ 103 ]have
been introduced to enable the processing of multiple datasets. Figure 9.4 illustrates
the framework of this model where the map phase transforms an input key/value pair
.k1; v 1/ into a list of intermediate key/value pairs Œ.k2; v 2/. The reduce function
aggregates the list of values Πv 2 associated with k2 and produces a list of values
Πv 3 which is also associated with k2. Note that inputs and outputs of both functions
belong to the same lineage (˛). Another pair of map and reduce functions produce