AnsweredAssumed Answered

Map reduce job statistics

Question asked by jacob_shaan on Dec 12, 2012
Latest reply on Dec 14, 2012 by jacob_shaan
I am running a 3 node M3 MAPR cluster with an additional node for CLDB, ZK, jobtracker and web-ui. I am running several map reduce jobs on the cluster. As the jobs finish, I was looking at their stats. There is a stat called 'CPU_MILLISECONDS' reported separately for map and reduce tasks and total - in MapReduce Framework Counters.

I was puzzled to see that while the job duration reported is 1 hr 2 mins and 52 sec, CPU_MILLISECONDS for the same job reported for map tasks is 4,058,920 ms (1 hr and 7 mins) and for reduce tasks is 4,043,570 ms (1 hr and 7 mins) - each of them being greater than the job duration itself. What does CPU\_MILLISECONDS in MapReduce Framework Counters refer to and how it is calculated?

I would appreciate if you could someone shed some light on it. Thanks in advance.