hadoop-mapreduce-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Neil Jonkers <neilod...@gmail.com>
Subject Re: total vcores per node containers in yarn
Date Sat, 18 Jul 2015 16:50:58 GMT

Also what scheduler are you using?
DefaultResourceCalculator only consider memory.


<div>-------- Original message --------</div><div>From: Shushant Arora <shushantarora09@gmail.com>
</div><div>Date:18/07/2015  4:18 PM  (GMT+02:00) </div><div>To: user@hadoop.apache.org
</div><div>Subject: Re: total vcores per node containers in yarn </div><div>
</div>Its hadoop 2.5.0.
Whats the logic of default using hardware detection. Say My node has 8 actual core and 32
virtual cores. Its taking 26 as value of vcores available of this node on RM UI.

On Sat, Jul 18, 2015 at 7:22 PM, Harsh J <harsh@cloudera.com> wrote:
What version of Apache Hadoop are you running? Recent changes have made YARN to auto-compute
this via hardware detection, by default (rather than the 8 default).

On Fri, Jul 17, 2015 at 11:31 PM Shushant Arora <shushantarora09@gmail.com> wrote:
In Yarn there is a setting to specify no of vcores that can be allocated to containers.


In my cluster's nodes yarn-site.xml, this property is not specified. But total vcores displayed
on RM's web page for the nodes are different than the default(8). Is there any other place
to control this number ?

View raw message