[
https://issues.apache.org/jira/browse/HAMA-304?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12919692#action_12919692
]
Edward J. Yoon commented on HAMA-304:
-------------------------------------
>> The implementation might not be very efficient for very large clusters, as a
>> list of groom server addresses is sent in each HeartBeat response. This is
>> something I think it should be improved sometime later.
This patch allows a peer to get a list of all other peers. so, all peers
collets and calculates the same result, after each peer executes its portion of
the loop a number of times. Right?
Only one peer can do that but +1, thinking we can improve Pi algorithm later.
Rest are looks great!
> Allow BSP peers to get a list of all other peers executing tasks of the same
> job
> --------------------------------------------------------------------------------
>
> Key: HAMA-304
> URL: https://issues.apache.org/jira/browse/HAMA-304
> Project: Hama
> Issue Type: New Feature
> Reporter: Filipe Manana
> Assignee: Filipe Manana
> Attachments: hama-304.patch
>
>
> Hi,
> Currently (as in the PiEstimator example), when a peer wants to send a
> message to other peer, it has to know in advance its hostname and port. This
> is likely fine for very small clusters, but not convenient for reasonably
> sized clusters.
> The following patch allows a peer to get a list of all other peers
> (collection of InetSocketAddress) so that it doesn't force a BSP job to know
> in advance the hostname and ports of the peers.
> The implementation might not be very efficient for very large clusters, as a
> list of groom server addresses is sent in each HeartBeat response. This is
> something I think it should be improved sometime later.
> Feedback welcome.
> cheers
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
