[
https://issues.apache.org/jira/browse/MAPREDUCE-326?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12829844#action_12829844
]
Tom White commented on MAPREDUCE-326:
-------------------------------------
> I don't see what this abstraction is buying you over using ByteBuffer and a
> Serializer that knows how to use it.
I can see a few benefits of a low-level binary API for framework developers:
* The new and old API could be written on top of it, effecting a cleaner
separation in the implementation.
* It would be easier to write powerful new APIs on top of the lower-level API.
E.g. MAPREDUCE-1183 would have more scope to deviate from the existing
(high-level) API than if it were written on top of the high-level API.
* It would be easier to optimize certain operations (and perhaps provide them
in a library): an identity mapper or reducer could just copy bytes; unused
fields in records could be skipped over (and not be deserialized).
* We could rewrite Pipes to pass the bytes read in the map/reduce directly to
the Pipes process. Currently Pipes has to do extra round trips. For example,
the Pipes Mapper in Java has to convert the records into Java types, then
serialize them when sending them to the Pipes process. Even in the case of
BytesWritable there is an extra copy that could be avoided.
> The lowest level map-reduce APIs should be byte oriented
> --------------------------------------------------------
>
> Key: MAPREDUCE-326
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-326
> Project: Hadoop Map/Reduce
> Issue Type: Improvement
> Reporter: eric baldeschwieler
>
> As discussed here:
> https://issues.apache.org/jira/browse/HADOOP-1986#action_12551237
> The templates, serializers and other complexities that allow map-reduce to
> use arbitrary types complicate the design and lead to lots of object creates
> and other overhead that a byte oriented design would not suffer. I believe
> the lowest level implementation of hadoop map-reduce should have byte string
> oriented APIs (for keys and values). This API would be more performant,
> simpler and more easily cross language.
> The existing API could be maintained as a thin layer on top of the leaner API.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
