Setting mapreduce.map.cpu.vcores and mapreduce.map.disk to zero caused a runTeraGenSort.sh to fail on a NEC 4 node system (24cores xeon, 192GB ram) as nearly 200 mappers were running on teragen which flooded MFS so bad it started to throw IOExceptions. One cluster finally passed but the other system failed teragen with this issue.
I recommend commenting out the values and relying on the defaults. I set the default values for these parameters so that they might be changed easily if need be.
Setting mapreduce.map.cpu.vcores and mapreduce.map.disk to zero caused a runTeraGenSort.sh to fail on a NEC 4 node system (24cores xeon, 192GB ram) as nearly 200 mappers were running on teragen which flooded MFS so bad it started to throw IOExceptions. One cluster finally passed but the other system failed teragen with this issue.
I recommend commenting out the values and relying on the defaults. I set the default values for these parameters so that they might be changed easily if need be.