You can also decommission taskTrackers when you know that you do not need it anymore.:
http://issues.apache.org/jira/browse/HADOOP-5643 thanks, dhruba On Wed, Jul 1, 2009 at 1:13 PM, Allen Wittenauer <[email protected]> wrote: > > > > On 7/1/09 12:51 PM, "Faisal Khan" <[email protected]> wrote: > In other words, the ability to schedule and run a single > > MapReduce job on a computing cluster that is also handling > > non-MapReduce job and vacating > > resources as soon as possible is very desirable for our high > > throughput community. > > Take a look at hod. This is essentially what it does, by letting PBS > (torque) schedule nodes to run MR and/or HDFS frameworks temporarily. > >
