Hi All, We have been using dual socket quad core machines for a while and have been running with 8 mappers 2 reducers. The rule of thumb we heard was slightly oversubscribe the number of cores but at Hadoop World several people said other things. Some new machines we are moving to have 2 socket 6 core machines with hyperthreading (24 'cores' if you count each hyperthreaded one as 2)
What are people doing for M/R ratios and number of M+R per machine at the moment? Thanks, Tom
