所有配置同hadoop一样。唯一多的地方是,在hadoop-site.xml 之中加入下列代码: <property> <name>mapred.map.pipeline</name> <value>true</value> <description>Pipeline Map Task Output . </description> </property> <property> <name>mapred.reduce.pipeline</name> <value>true</value> <description>Pipeline Reduce Task Output A pipelined reduce task will send its output directly to the map task in the subsequent job, if any. </description> </property> <property> <name>mapred.snapshot.frequency</name> <value>0.1</value> <description>Snapshot frequency Determines how often we perform snapshots based on job progress. A freq=0.1 will perform snapshots at 10% progress intervals (e.g., 10%, 20%, ..., 100%). </description> </property> <property> <name>mapred.job.input.snapshots</name> <value>false</value> <description>Job Input is Snapshot If a job takes snapshot results as input then this parameter must be set to true. </description> </property>