These are all very predictable, I think, since they are the sum of a lot of small operations that don't vary much.
On Wed, Apr 14, 2010 at 2:16 PM, Byung-Gon Chun <bgc...@gmail.com> wrote: > > This may be too dumb but if your goal is to run a mapper/reducer that > > takes a long time sometimes, just have then run a mapper that sleeps a > > long time some times? a random number of milliseconds with wide range? > > > > > Yes. But I'd like to find realistic workload. How about classification or > clustering examples?
