On Thursday 29 September 2011 02:56 PM, Tanweiguo wrote: > s we may get better time achievement rat
When we process a short running job on Hadoop, framework overhead is more than job processing time. When MR executes a job, as we know TT needs to spawn atleast one process (atleast map jvm, suppose job does not have any reducers). But when you execute the same short running job as a non hadoop job, there won't be no framework overhead time. I feel it could be the reason in your experiment. Watch out for more comments. -- /Thanks&Regards,/ /Bh.V.S.Kamesh/
