[
https://issues.apache.org/jira/browse/MAPREDUCE-64?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12767693#action_12767693
]
Todd Lipcon commented on MAPREDUCE-64:
--------------------------------------
Thanks for those great diagrams - they really helped me understand things much
better! A picture is worth 1000 lines of code or something :)
I applied your patch just now and ran it through clover for coverage analysis.
Here are a couple things I think we should cover before committing:
- We don't current run any tests with job.getCompressMapOutput returning true.
This caused an issue or two with the shuffle in the past, so we should get at
least one test that uses a codec.
- Since we're using the Local Runner for these tests, it's all a single
partition. This is probably OK, since I imagine other tests throughout Hadoop
exercise those paths (I'm only looking at coverage from TestMapCollection here)
- Line 1097 ("if (bufindex + headbytelen < avail) {" in void reset()) is always
true in our tests. We should get a test case to exercise the other half of this
branch.
- Line 1365 (kvstart >= kvend ternary in sortAndSpill) is always true. Should
exercise the other half of that.
On a code level, one more thing I noticed - can you put in a small comment
describing the synchronization policy for the various offsets? Those used to be
volatile and now they're under a lock, so it should be good to note that in the
code.
I'll try to get a chance to run some basic benchmarks later this week.
> Map-side sort is hampered by io.sort.record.percent
> ---------------------------------------------------
>
> Key: MAPREDUCE-64
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-64
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Reporter: Arun C Murthy
> Assignee: Chris Douglas
> Attachments: M64-0.patch, M64-0i.png, M64-1.patch, M64-1i.png,
> M64-2.patch, M64-2i.png, M64-3.patch
>
>
> Currently io.sort.record.percent is a fairly obscure, per-job configurable,
> expert-level parameter which controls how much accounting space is available
> for records in the map-side sort buffer (io.sort.mb). Typically values for
> io.sort.mb (100) and io.sort.record.percent (0.05) imply that we can store
> ~350,000 records in the buffer before necessitating a sort/combine/spill.
> However for many applications which deal with small records e.g. the
> world-famous wordcount and it's family this implies we can only use 5-10% of
> io.sort.mb i.e. (5-10M) before we spill inspite of having _much_ more memory
> available in the sort-buffer. The word-count for e.g. results in ~12 spills
> (given hdfs block size of 64M). The presence of a combiner exacerbates the
> problem by piling serialization/deserialization of records too...
> Sure, jobs can configure io.sort.record.percent, but it's tedious and
> obscure; we really can do better by getting the framework to automagically
> pick it by using all available memory (upto io.sort.mb) for either the data
> or accounting.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
