On Thu, Aug 27, 2009 at 12:51 PM, Rares Vernica<[email protected]> wrote: > > Another difference is the fact that the keys for Job 1 are Text, while the > keys for Job 2 are IntWritable (words are converted to integers).
I changed the key type of Job 2 from IntWritable to Text and the merge phase performance is the same. The Map output bytes increased slightly. It seems that merging data from disk only is faster than merging data from memory and disk... Cheers, Rares Vernica
