[
https://issues.apache.org/jira/browse/YARN-1290?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13794817#comment-13794817
]
Wei Yan commented on YARN-1290:
-------------------------------
Thanks, [~sandyr].
Here we resort all nodes before assignment in each round, instead of
maintaining a PriorityQueue which may bring more cost (as each node-related
update causes re-insert operation).
The performance is also related to the node heartbeat interval. If lots of
heartbeats come in, the continuous scheduling thread has less resource requests
to handle.
> Let continuous scheduling achieve more balanced task assignment
> ---------------------------------------------------------------
>
> Key: YARN-1290
> URL: https://issues.apache.org/jira/browse/YARN-1290
> Project: Hadoop YARN
> Issue Type: Improvement
> Reporter: Wei Yan
> Assignee: Wei Yan
> Attachments: main.pdf, YARN-1290.patch, YARN-1290.patch,
> YARN-1290.patch
>
>
> Currently, in continuous scheduling (YARN-1010), in each round, the thread
> iterates over pre-ordered nodes and assigns tasks. This mechanism may
> overload the first several nodes, while the latter nodes have no tasks.
> We should sort all nodes according to available resource. In each round,
> always assign tasks to nodes with larger capacity, which can balance the load
> distribution among all nodes.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
