[
https://issues.apache.org/jira/browse/MAPREDUCE-2841?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13169992#comment-13169992
]
Todd Lipcon commented on MAPREDUCE-2841:
----------------------------------------
I chatted recently with the MR guys over at Facebook. They have another
implementation here which they've been working on that gives similar gains,
while staying all in java. The approach is something like the following:
- map output collector collects into small buffers, each sized something close
to L3 cache
- when any buffer is full, sort it but don't spill it
- when enough buffers are collected to fill io.sort.mb, merge them from memory
to disk
This fixes the cache locality issues that everyone has identified, but doesn't
require native code. It's up on their github here:
https://github.com/facebook/hadoop-20/blob/master/src/mapred/org/apache/hadoop/mapred/BlockMapOutputBuffer.java
Maybe Yongqiang can comment more on this approach? Dmytro has given permission
offline for us to work from the code on their github and contribute it on trunk
(they may not have time to contribute it in the nearterm)
I think a short term goal for this area we could attack would be to make the
map output collector implementation pluggable. Then people can experiment more
freely with different collector implementations. I don't have time for it -
just throwing it out there as a thought.
> Task level native optimization
> ------------------------------
>
> Key: MAPREDUCE-2841
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-2841
> Project: Hadoop Map/Reduce
> Issue Type: Improvement
> Components: task
> Environment: x86-64 Linux
> Reporter: Binglin Chang
> Assignee: Binglin Chang
> Attachments: MAPREDUCE-2841.v1.patch, MAPREDUCE-2841.v2.patch,
> dualpivot-0.patch, dualpivotv20-0.patch
>
>
> I'm recently working on native optimization for MapTask based on JNI.
> The basic idea is that, add a NativeMapOutputCollector to handle k/v pairs
> emitted by mapper, therefore sort, spill, IFile serialization can all be done
> in native code, preliminary test(on Xeon E5410, jdk6u24) showed promising
> results:
> 1. Sort is about 3x-10x as fast as java(only binary string compare is
> supported)
> 2. IFile serialization speed is about 3x of java, about 500MB/s, if hardware
> CRC32C is used, things can get much faster(1G/s).
> 3. Merge code is not completed yet, so the test use enough io.sort.mb to
> prevent mid-spill
> This leads to a total speed up of 2x~3x for the whole MapTask, if
> IdentityMapper(mapper does nothing) is used.
> There are limitations of course, currently only Text and BytesWritable is
> supported, and I have not think through many things right now, such as how to
> support map side combine. I had some discussion with somebody familiar with
> hive, it seems that these limitations won't be much problem for Hive to
> benefit from those optimizations, at least. Advices or discussions about
> improving compatibility are most welcome:)
> Currently NativeMapOutputCollector has a static method called canEnable(),
> which checks if key/value type, comparator type, combiner are all compatible,
> then MapTask can choose to enable NativeMapOutputCollector.
> This is only a preliminary test, more work need to be done. I expect better
> final results, and I believe similar optimization can be adopt to reduce task
> and shuffle too.
--
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
