I config them in mapred-site.xml like below, I set them less then 1000 for
the normalization as you said:
"
<property>
<name>yarn.app.mapreduce.am.resource.mb</name>
<value>900</value>
</property>
<property>
<name>mapreduce.map.memory.mb</name>
<value>900</value>
</property>
<property>
<name>mapreduce.reduce.memory.mb</name>
<value>900</value>
</property>
"
Then I run just one map, as you said there are 2 contained will be
launched, one for A/M master, the other for map task.
However, the 2 container cost 4G memory which I see from yarn UI interface.
2013/4/24 Zhijie Shen <[email protected]>
> Do you mean the memory assigned for the container of M/R's AM? Did you set
> ContainerLaunchContext.setResource?
>
> AFAIK, by default, yarn.scheduler.minimum-allocation-mb = 1024 and
> yarn.app.mapreduce.am.resource.mb
> = 1536. So, M/R job will request 1536 for its AM, but Yarn's scheduler will
> normalize the request to 2048, which is no less than 1536, and is multiple
> times of the min allocation.
>
>
> On Tue, Apr 23, 2013 at 8:43 AM, 牛兆捷 <[email protected]> wrote:
>
> > I am using 2.0.3-alpha, I don't set the map memory capacity explicitly,
> > then "resourceCapacity.setMemory" should set the default memory request
> to
> > 1024mb,
> > However 2048 Memory is assigned to this container.
> >
> > Why it does like this?
> >
> > --
> > *Sincerely,*
> > *Zhaojie*
> > *
> > *
> >
>
>
>
> --
> Zhijie Shen
> Hortonworks Inc.
> http://hortonworks.com/
>
--
*Sincerely,*
*Zhaojie*
*
*
