Details

    • Type: Improvement Improvement
    • Status: Resolved
    • Priority: Major Major
    • Resolution: Fixed
    • Affects Version/s: None
    • Fix Version/s: 3.0.0
    • Component/s: task
    • Labels:
      None
    • Environment:

      x86-64 Linux/Unix

    • Hadoop Flags:
      Reviewed
    • Release Note:
      Hide
      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.
      Show
      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

      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.

      1. DESIGN.html
        42 kB
        Binglin Chang
      2. dualpivot-0.patch
        5 kB
        Chris Douglas
      3. dualpivotv20-0.patch
        4 kB
        Chris Douglas
      4. fb-shuffle.patch
        76 kB
        Todd Lipcon
      5. hadoop-3.0-mapreduce-2841-2014-7-17.patch
        3.50 MB
        Sean Zhong
      6. MAPREDUCE-2841.v1.patch
        180 kB
        Binglin Chang
      7. MAPREDUCE-2841.v2.patch
        190 kB
        Binglin Chang
      8. micro-benchmark.txt
        13 kB
        Todd Lipcon
      9. MR-2841benchmarks.pdf
        213 kB
        Todd Lipcon
      10. mr-2841-merge.txt
        2.72 MB
        Todd Lipcon
      11. mr-2841-merge-2.txt
        2.70 MB
        Sean Zhong
      12. mr-2841-merge-3.patch
        2.73 MB
        Sean Zhong
      13. mr-2841-merge-4.patch
        2.68 MB
        Sean Zhong

        Issue Links

          Activity

          Binglin Chang created issue -
          Binglin Chang made changes -
          Field Original Value New Value
          Assignee Binglin Chang [ decster ]
          Binglin Chang made changes -
          Status Open [ 1 ] Patch Available [ 10002 ]
          Binglin Chang made changes -
          Attachment MAPREDUCE-2841.v1.patch [ 12490790 ]
          Chris Douglas made changes -
          Attachment dualpivot-0.patch [ 12491994 ]
          Attachment dualpivotv20-0.patch [ 12491995 ]
          Binglin Chang made changes -
          Attachment MAPREDUCE-2841.v2.patch [ 12492208 ]
          Binglin Chang made changes -
          Link This issue relates to MAPREDUCE-3246 [ MAPREDUCE-3246 ]
          Binglin Chang made changes -
          Link This issue relates to MAPREDUCE-3247 [ MAPREDUCE-3247 ]
          Binglin Chang made changes -
          Attachment DESIGN.html [ 12512978 ]
          Dong Yang made changes -
          Link This issue relates to MAPREDUCE-1270 [ MAPREDUCE-1270 ]
          chwcrazy made changes -
          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/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.




          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.




          chwcrazy made changes -
          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.




          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.




          chwcrazy made changes -
          Environment x86-64 Linux x86-64 Linux/Unix
          Todd Lipcon made changes -
          Attachment fb-shuffle.patch [ 12638882 ]
          Todd Lipcon made changes -
          Assignee Binglin Chang [ decster ] Sean Zhong [ clockfly ]
          Todd Lipcon made changes -
          Link This issue is related to MAPREDUCE-5962 [ MAPREDUCE-5962 ]
          Sean Zhong made changes -
          Status Patch Available [ 10002 ] Open [ 1 ]
          Sean Zhong made changes -
          Todd Lipcon made changes -
          Link This issue is related to HADOOP-10855 [ HADOOP-10855 ]
          Todd Lipcon made changes -
          Attachment micro-benchmark.txt [ 12665471 ]
          Todd Lipcon made changes -
          Attachment MR-2841benchmarks.pdf [ 12666128 ]
          Todd Lipcon made changes -
          Attachment mr-2841-merge.txt [ 12666832 ]
          Todd Lipcon made changes -
          Status Open [ 1 ] Patch Available [ 10002 ]
          Sean Zhong made changes -
          Status Patch Available [ 10002 ] In Progress [ 3 ]
          Sean Zhong made changes -
          Attachment mr-2841-merge-2.txt [ 12666979 ]
          Sean Zhong made changes -
          Status In Progress [ 3 ] Patch Available [ 10002 ]
          Sean Zhong made changes -
          Attachment mr-2841-merge-3.patch [ 12666986 ]
          Sean Zhong made changes -
          Status Patch Available [ 10002 ] In Progress [ 3 ]
          Sean Zhong made changes -
          Status In Progress [ 3 ] Patch Available [ 10002 ]
          Sean Zhong made changes -
          Status Patch Available [ 10002 ] In Progress [ 3 ]
          Sean Zhong made changes -
          Status In Progress [ 3 ] Patch Available [ 10002 ]
          Sean Zhong made changes -
          Status Patch Available [ 10002 ] Open [ 1 ]
          Sean Zhong made changes -
          Attachment mr-2841-merge-4.patch [ 12667067 ]
          Sean Zhong made changes -
          Status Open [ 1 ] Patch Available [ 10002 ]
          Todd Lipcon made changes -
          Status Patch Available [ 10002 ] Resolved [ 5 ]
          Hadoop Flags Reviewed [ 10343 ]
          Release Note Task level native optimization 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.
          Fix Version/s 3.0.0 [ 12320355 ]
          Resolution Fixed [ 1 ]
          Allen Wittenauer made changes -
          Link This issue relates to MAPREDUCE-6106 [ MAPREDUCE-6106 ]

            People

            • Assignee:
              Sean Zhong
              Reporter:
              Binglin Chang
            • Votes:
              4 Vote for this issue
              Watchers:
              70 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved:

                Development