You should try to utilize maximum of your RAM to map tasks. You can do that by changing the parameters at mapred-site.xml file. add the below statements to set 6.5 GB of RAM for map tasks <property> <name>mapred.map.child.java.opts</name> <value>-Xmx6500m</value> <description>heap size for map tasks </description> </property>
On Mon, Oct 21, 2013 at 11:33 AM, Puneet Jain <puneetdabu...@gmail.com>wrote: > Hello: > > I have a graph of size 2B edges, 200M vertices. I am trying to compute > pagerank of this graph. I have 10 node cluster, 8G memory. > > I am unable to run giraph job successfully. My question is more generic: > > 1. How should one choose right amount of heap space? .. is there a way of > telling giraph job not to pick too much computation, if it can't handle in > memory? > > 2. Does out-of-core message feature functions as expected ? > > > > -- > --Puneet >