You can find out the reason from the JT logs (eg: memory/timeout restrictions) and adjust the timeout - mapred.task.timeout or the memory parameters accordingly.Refer http://hadoop.apache.org/common/docs/r0.20.0/cluster_setup.html Cheers, /R
On 1/29/10 12:22 PM, "john li" <lij...@gmail.com> wrote: when hadoop running multi jobs concurrently, that is when hadoop is busy, always have killed tasks in some jobs, although the jobs success finally. anybody tell me why? -- Regards Junyong