6.What should you do?
Each node in your Hadoop cluster, running YARN, has 64GB memory and 24 cores.
Your yarn.site.xml has the following configuration:

yarn.nodemanager.resource.memory-mb
32768


yarn.nodemanager.resource.cpu-vcores
12

You want YARN to launch no more than 16 containers per node. What should you do?

A.
Modify yarn-site.xml with the following property:
yarn.scheduler.minimum-allocation-mb
2048
B.
Modify yarn-sites.xml with the following property:
yarn.scheduler.minimum-allocation-mb
4096
C.
Modify yarn-site.xml with the following property:
yarn.nodemanager.resource.cpu-vccores
D.
No action is needed: YARN’s dynamic resource allocation automatically optimizes the node
memory and cores

问题:
      Hadoop集群的每台节点内存64G,CPU 24核。当前一个节点可分配的物理内存总量是32768M(32G),可分配的虚拟cpu总个数是12。
   那么yarn在每个节点上,发起不超过16个容器,我们应该怎样配置？

分析:A

   yarn.scheduler.minimum-allocation-mb：最小可申请内存量，默认是1024,我们设置2048M,见如下计算公式

       Maximum memory YARN can utilize on the node
    ————————————————————————————————————————————— = minimum memory per container
       Number of containers
    
配置文件                     配置项名称                                  配置项值
yarn-site.xml        yarn.nodemanager.resource.memory-mb        = Containers个数* 每个Container内存
yarn-site.xml        yarn.scheduler.minimum-allocation-mb        = 每个Container内存
yarn-site.xml        yarn.scheduler.maximum-allocation-mb        = Containers个数* 每个Container内存
mapred-site.xml        mapreduce.map.memory.mb        = 每个Container内存
mapred-site.xml        mapreduce.reduce.memory.mb        = 2 * 每个Container内存
mapred-site.xml        mapreduce.map.java.opts        = 0.8 * 每个Container内存
mapred-site.xml        mapreduce.reduce.java.opts        = 0.8 * 2 * 每个Container内存
yarn-site.xml (check)        yarn.app.mapreduce.am.resource.mb        = 2 * 每个Container内存
yarn-site.xml (check)        yarn.app.mapreduce.am.command-opts        = 0.8 * 2 * 每个Container内存

http://developer.51cto.com/art/201401/426610.htm