Configuring Fair scheduler

[Biju N <bi...@gmail.com>]

*Environment: *hadoop version 2.6 / Ubuntu

I am trying to run the map reduce example program "pi" on a cluster with
one RM and one NM. The nodes are VMs with 4 processors and 4 MB memory and
the hypervisor hosting the VMs have 8 cores and 16 GB memory. When I run
the job with the following properties in yarn-site.xml the job doesn't get
scheduled. But when the "memory-mb" property is changed to 4096 ( which is
the total memory on the node ) the job completes successfully. Note that
this properties works fine in hadoop 2.2. Also only the default queue is
used and this is the only job on the cluster.

  <property>
    <name>yarn.nodemanager.resource.memory-mb</name>
    <value>3072</value>
  </property>

  <property>
    <name>yarn.nodemanager.resource.cpu-vcores</name>
    <value>3</value>
  </property>

<property>
    <name>yarn.resourcemanager.scheduler.class</name>

<value>org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairScheduler</value>
  </property>

The question is what is the calculation which need to be used to set the
following properties in yarn-site.xml

yarn.nodemanager.resource.memory-mb
yarn.nodemanager.resource.cpu-vcores

Also how to derive the values for the following properties in the
fair-scheduler.xml

    <minResources>nnn mb,0vcores</minResources>
    <maxResources>nnn mb,2vcores</maxResources>

Any help on this is much appreciated.

Thanks,
Biju