[ 
https://issues.apache.org/jira/browse/GIRAPH-259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13427285#comment-13427285
 ] 

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

As long as we document what's the way of execution, having MasterCompute being 
executed before or after Vertex computes essentially differs just in the first 
super step - whether algorithm starts with master or vertex compute. From what 
I see, now it goes like this:
input superstep -> master compute 0 -> vertex compute 0 -> master compute 1 -> 
vertex compute 1 -> ...
One thing is that master can't actually give any aggregator value to vertices 
in superstep 0 since vertices didn't get a chance to call registerAggregator 
before that. I guess the reason we ask aggregators to be registered on the user 
is so we wouldn't have to send aggregator class to workers (or in current 
implementation write it to Zookeeper)?

Here is the problem with current approach: say we have a regular (not 
persistent) sum aggregator and master and vertices make the following changes 
during computation:
||superstep||compute||value it sees||change||
|1|master|0|+1|
|1|vertex|1|+5|
|2|master|6|+10|
|2|vertex|15 or 16?| |

We can just define one of those, but my opinion is that it would be clearer if 
vertices get executed first:
||superstep||compute||value it sees||change||
|1|vertex|0|+5|
|1|master|5|+1|
|2|vertex|6|+50|
|2|master|50|+10|
|3|vertex|60| |
And if we want to start with master compute we can have master compute 
superstep -1 for example.
                
> 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