[MAPREDUCE-2841] Task level native optimization

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.0.0-alpha1
• Component/s: task

Environment

x86-64 Linux/Unix

• Hadoop Flags: Reviewed

Release Note

Adds a native implementation of the map output collector. The native library will build automatically with -Pnative. Users may choose the new collector on a job-by-job basis by setting mapreduce.job.map.output.collector.class=org.apache.hadoop.mapred.
nativetask.NativeMapOutputCollectorDelegator in their job configuration. For shuffle-intensive jobs this may provide speed-ups of 30% or more.

• Tags: optimization task
• Assignee: Sean Zhong
• Reporter: Binglin Chang
• Created: 13/Aug/11 07:29
• Updated: 19/Oct/17 05:54
• Resolved: 13/Sep/14 01:47

Description

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/

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.

Issue Links

• duplicates: MAPREDUCE-1270 Hadoop C++ Extention
• duplicates: MAPREDUCE-2446 HCE 2.0
• is depended upon by: HIVE-17498 Does hive have mr-nativetask support refer to MAPREDUCE-2841
• is related to: MAPREDUCE-6985 MapReduce native optimization does not work properly due to a shuffle error (LocalJobRunner)
• is related to: MAPREDUCE-5962 Support CRC32C in IFile
• is related to: HADOOP-10855 Allow Text to be read with a known length
• relates to: MAPREDUCE-6106 hadoop-mapreduce-client-nativetask fails to compile on OS X
• relates to: MAPREDUCE-3247 Add hash aggregation style data flow and/or new API
• relates to: MAPREDUCE-3246 Make Task extensible to support modifications of Task or even alternate programming paradigms
• relates to: MAPREDUCE-1270 Hadoop C++ Extention

Sub-Tasks

• native-task test logs should not write to console: Open