AnsweredAssumed Answered

What can cause task startup delay?

Question asked by jeffcb on Aug 5, 2011
Latest reply on Aug 9, 2011 by amit
I ran the pi test with 16 tasks, and 10B samples each.  Each of my 4 nodes is allowed to run 8 map tasks (more than the 4 I'm asking for).  With open source hadoop, this test runs consistently in 520 seconds.  With MapR, it varies from 540 to about 600 seconds.  I found out that 3 of the nodes start 4 map tasks each immediately and simultaneously, but the third node starts its 4 map tasks after a delay.  This delay ranges from 5 seconds to 2.5 minutes for different runs.  The machine is idle during the delay. Rebooting the cluster doesn't change anything.  Of course, the third node finishes its tasks later than the other nodes and causes the increase in overall execution time.  As far as I know the third node is configured the same as the others.

What could cause this random delay?