[
https://issues.apache.org/jira/browse/MESOS-3401?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14909348#comment-14909348
]
haosdent commented on MESOS-3401:
---------------------------------
This issue remind me of node labels in Hadoop YARN.
http://www.slideshare.net/Hadoop_Summit/node-labels-in-yarn-49792443 And Yahoo
use this feature to implement Deep Learning in YARN
http://yahoohadoop.tumblr.com/post/129872361846/large-scale-distributed-deep-learning-on-hadoop
> Add labels to Resources
> -----------------------
>
> Key: MESOS-3401
> URL: https://issues.apache.org/jira/browse/MESOS-3401
> Project: Mesos
> Issue Type: Improvement
> Components: slave
> Reporter: Adam B
> Assignee: Greg Mann
> Labels: mesosphere, resources
>
> Similar to how we have added labels to tasks/executors (MESOS-2120), and even
> FrameworkInfo (MESOS-2841), we should extend Resource to allow arbitrary
> key/value pairs.
> This could be used to specify that a cpu resource has a certain speed, that a
> disk resource is SSD, or express any other metadata about a built-in or
> custom resource type. Only the scalar quantity will be used for determining
> fair share in the Mesos allocator. The rest will be passed onto frameworks as
> info they can use for scheduling decisions.
> This would require changes to how the slave specifies its `--resources`
> (probably as json), how the slave/master reports resources in its web/json
> API, and how resources are offered to frameworks.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
