The number of tasks (map/reduce) would also depend on how many cores you have on your machine and not just the memory available. Its generally a good idea to have 1-2X more tasks than the number of cores. For eg, for a 4 core machine you could have a total of 4-6 tasks (map+reduce) on a single machine. Remember that TT and DN each is a separate process.
Also, I was told at a conference by a presenter from Cloudera that Map:Reduce ratio should ideally be 4:3. But you could decide that based on your requirements. Thanks, Prashant On Thu, Nov 17, 2011 at 8:55 AM, Keren Ouaknine <[email protected]> wrote: > Hello, > > I am running on machines with 4G, out of which two are allocated for > running the OS in memory. > It leaves me with 2G, I will be using 1.5 to run the mappers and reducers > (machines are dedicated) and 0.5 to run the box. > Running the 17 scripts, I was wondering whether I should configure: > - 2 mappers and one reducer with 516M each > - 6 mappers and one reducer with 256M each > > Thanks, > Keren > > -- > Keren Ouaknine > Cell: +972 54 2565404 > Web: www.kereno.com >
