[ https://issues.apache.org/jira/browse/HAMA833?page=com.atlassian.jira.plugin.system.issuetabpanels:commenttabpanel&focusedCommentId=13855522#comment13855522
]
Ilias Kapouranis edited comment on HAMA833 at 12/23/13 9:36 AM:

If you want to find the min or max (for example) value from the vertices and continue with
the ones that satisfy the condition, having the nodes being aggregated at the superstep they
vote to halt hides some errors. I came up with the problem and we discussed it a bit at the
user forum.
link : http://mailarchives.apache.org/mod_mbox/hamauser/201312.mbox/%3C20131220174851.Horde.IOFgOj1UIAoxM5AG2FmErQ2@webmail.auth.gr%3E
Specifically:
Your algorithm has 2 supersteps for each of its' iteration.
 At the first superstep assigns are done based on the aggregated values and messages are
send.
 At the second nodes start voting to halt based on the messages received.
Now the problem here is that the nodes which voted to halt will be aggregated and will mess
up the result for the next iterations.
Having the option to exclude them from the aggregation phase in which they voted to halt gives
better control.
was (Author: ikapoura):
If you want to find the min or max (for example) value from the vertexes and continue with
the ones that satisfy the condition, having the nodes being aggregated at the superstep they
vote to halt hides some errors. I came up with the problem and we discussed it a bit at the
user forum.
link : http://mailarchives.apache.org/mod_mbox/hamauser/201312.mbox/%3C20131220174851.Horde.IOFgOj1UIAoxM5AG2FmErQ2@webmail.auth.gr%3E
Specifically:
Your algorithm has 2 supersteps for each of its' iteration.
 At the first superstep assigns are done based on the aggregated values and messages are
send.
 At the second nodes start voting to halt based on the messages received.
Now the problem here is that the nodes which voted to halt will be aggregated and will mess
up the result for the next iterations.
Having the option to exclude them from the aggregation phase in which they voted to halt gives
better control.
> Add more finish states for vertices
> 
>
> Key: HAMA833
> URL: https://issues.apache.org/jira/browse/HAMA833
> Project: Hama
> Issue Type: Improvement
> Components: graph
> Affects Versions: 0.6.3
> Reporter: Anastasis Andronidis
> Assignee: Anastasis Andronidis
> Priority: Minor
> Labels: features
> Fix For: 0.7.0
>
>
> We should handle more cases on the vertices, like:
> 1) voteToStop() : Immediately stop the vertex compute and suppress any further calculations
on top of that. (e.g. aggregation)
> 2) voteToTerminate(): Immediately stop the vertex compute, suppress any further calculations
on top of that and deactivate the vertex so even if any message reaches it, will not come
alive.
> Any comments?

This message was sent by Atlassian JIRA
(v6.1.5#6160)
