Dne 11.6.2013 2:46, Grandl Robert napsal(a):
I do not think that this is possible in hadoop but its pretty common in other mapreduce software. They can even chain map-reduce-map without writing to disk and moving it across network to different node.Hi,It is possible to run shuffle phase in memory ? I mean, the mapper to write the intermediate output in memory only and the reducer to read it from there. So no disk involved for intermediate output.
