For specific max-heap sizes, you have to pass the value as a java vm argument. See http://avricot.com/blog/index.php?post/2010/05/03/Get-started-with-java-JVM-memory-(heap%2C-stack%2C-xss-xms-xmx-xmn...) for a good view on things with JVM and memory.
An example for specific heap size options to JVM: # 1024 MB for DN HADOOP_DATANODE_OPTS="$HADOOP_DATANODE_OPTS -Xmx1024m" # 4 GB for NN HADOOP_NAMENODE_OPTS="$HADOOP_NAMENODE_OPTS -Xmx4g" On Mon, Aug 1, 2011 at 11:41 PM, Kai Ju Liu <ka...@tellapart.com> wrote: > Hi. I'm trying to tweak heap sizes for the Hadoop daemons, i.e. > namenode/datanode and jobtracker/tasktracker. I've tried setting > HADOOP_NAMENODE_HEAPSIZE, HADOOP_DATANODE_HEAPSIZE, and so on in > hadoop-env.sh, but the heap size remains at the default of 1,000MB. > > In the cluster setup documentation, I see references to setting > HADOOP_NAMENODE_OPTS, HADOOP_DATANODE_OPTS, and so on in hadoop-env.sh. Is > this the proper way to set heap sizes, and if so, what is the proper syntax > within the OPTS values? Thanks! > > Kai Ju Liu > -- Harsh J
