Map Reduce Performance improvement by using simple file copy rather than using HTTP, when mapper and reducer runs on same TT

[Kamesh]

Hi All,
In the current Implementation of MR, suppose say Mapper and Reducer are 
running in the same Task Tracker. After completion of Map task, 
Reducer(s) copy the mapper output using HTTP. But I feel when reducer 
runs on a same TT, where mapper had run, using simple file copy to fetch 
the mapper output gives better performance.

Any comments on this?

If this is valid, I will raise it in JIRA.
--

Thanks&Regards,
Bh.V.S.Kamesh