[
https://issues.apache.org/jira/browse/GIRAPH-192?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13396304#comment-13396304
]
Jakob Homan commented on GIRAPH-192:
------------------------------------
Hey Jan. Had a look at the patch. A few questions:
* What's the use case for the overwrite aggregator? It's nondeterministic in
operation (particularly in a distributed environment), so I'm having trouble
seeing why it's needed?
* Rather than separate Int/Long and Float/Double aggregators, perhaps we can
just have long and float, respectively, and rely on the greater size/precision?
Each aggregator will take twice as much memory, but there should be relatively
few of them so this may not be so bad. Thoughts?
* It seems like it should be possible to parameterize these classes on pairs of
both the Java type and corresponding Writable version (ie <Long,
LongWritable>). Not great, but might lead to less code. Is this viable?
* The new aggregators will require unit tests and we may as well take the
opportunity to add unit tests for the existing ones (if not, at least add a
newbie JIRA for adding them).
* What happens when we overflow on, for instance, product aggregator with an
int? This should be tested as well.
> Move aggregators to a seperate sub-package
> ------------------------------------------
>
> Key: GIRAPH-192
> URL: https://issues.apache.org/jira/browse/GIRAPH-192
> Project: Giraph
> Issue Type: Improvement
> Components: examples
> Affects Versions: 0.2.0
> Reporter: Jan van der Lugt
> Assignee: Jan van der Lugt
> Priority: Minor
> Fix For: 0.2.0
>
> Attachments: GIRAPH-192.patch
>
> Original Estimate: 2h
> Remaining Estimate: 2h
>
> Since aggregators will be re-used throughout many projects and algorithms, it
> makes sense to implement the most common ones in a separate sub-package. This
> will reduce the time required for users when they implement their projects
> based on Giraph, because the required aggregators are already in place. I
> implemented the following ones:
> for int/long/float/double: min, max, product, sum, overwrite
> for boolean: and, or, overwrite
> Most of them speak for themselves, except for the overwrite one. This
> aggregator simply overwrites the stored value when a new value is aggregated.
> This is useful when one node is in some way a master node (for example a
> source node in an routing algorithm), and this node wants to broadcast a
> value to all other nodes.
> Attached is a patch against trunk implementing the aggregators and patching
> some existing files so they use the .aggregators package instead of the
> .examples one.
--
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
