[ 
https://issues.apache.org/jira/browse/GIRAPH-259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13427958#comment-13427958
 ] 

Maja Kabiljo commented on GIRAPH-259:
-------------------------------------

I thought giving an example would make things clear, but I see now I should 
have made it longer in the beginning, sorry about that.

In current patch there is no more setAggregatedValue available to user. I 
thought of the same thing you said about having different options for vertices 
and master, but kind of wanted to make aggregators at least work first and then 
raise that issue. But now that it's out, we can discuss the whole design more.

I am still not sure what exactly do you mean when you say that MasterCompute 
should handle aggregators completely. Do you think just the resetting part or 
some other things too? I can see the benefit of giving user resetting 
aggregator option without him needing to do anything. And if he wants it to 
work in any different way he can still define the persistent one and do 
whatever he wants with it (as long as we provide different AggregatorUsage 
interfaces for vertex and master). But if the master doesn't touch the value, 
we should send out the one aggregated in the previous super step.

It would for sure be nice to get rid of registerAggregator and useAggregator on 
workers. Even though that would mean that if the user wants to use aggregators, 
it's not enough to implement Vertex class. If we are worried about performance, 
aggregators could internally have a flag whether their value was updated or 
not, so they don't send the value to master if it wasn't updated. Other thing 
is that we would need to introduce aggregator types, so master would send it 
and worker would know which class to instantiate.

Just as a note, we shouldn't rely on aggregators being small anymore, we have 
an application in which we need big aggregators and a lot of them (which was 
the reason for raising GIRAPH-273).
                
> TestBspBasic.testBspPageRank is broken
> --------------------------------------
>
>                 Key: GIRAPH-259
>                 URL: https://issues.apache.org/jira/browse/GIRAPH-259
>             Project: Giraph
>          Issue Type: Bug
>            Reporter: Maja Kabiljo
>            Assignee: Maja Kabiljo
>         Attachments: GIRAPH-259-1.patch, GIRAPH-259-2.patch, 
> GIRAPH-259-3.patch
>
>
> Test crashes on line 152 in class SimplePageRankVertex in distributed mode.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Reply via email to