[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16265109#comment-16265109 ] Li Yuanjian commented on SPARK-2926: Yes, only the reduce stage. You're right, I shouldn't only pay attention to the final stage. I rearrange all messages of screenshot below: ||test name||map stage shuffle write||reduce stage shuffle read||map stage duration||reduce stage duration||total job time||code base|| |Test Round1|654.4 MB|802.6MB|3.6min|2.0min|5.5min|master| |Test Round1|654.4 MB|714.0MB|3.4min|9s|3.5min|SPARK-2926| |Test Round2: Add more pressure for SortShuffleReader by coalesce|654.4MB|654.4MB|2.6min|20min|22min|master| |Test Round2: Add more pressure for SortShuffleReader by coalesce|654.4MB|654.4MB|3.7min|1.4min|5.1min|SPARK-2926| |Test Round3: Test file spill scenario of sort shuffle reader|142.6MB|142.6MB|26s|16min|16min|master| |Test Round3: Test file spill scenario of sort shuffle reader|142.6MB|142.6MB|21s|13s|34s|SPARK-2926| |Test Round3: Test file spill scenario of sort shuffle reader|142.6MB|142.6MB|22s|25s|47s|SPARK-2926(code change for force spill to disk)| |Test Round3: Test file spill scenario of sort shuffle reader|142.6MB|142.6MB|22s|29s|51s|SPARK-2926(code change for force spill to disk)| > Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle > -- > > Key: SPARK-2926 > URL: https://issues.apache.org/jira/browse/SPARK-2926 > Project: Spark > Issue Type: Improvement > Components: Shuffle >Affects Versions: 1.1.0 >Reporter: Saisai Shao >Assignee: Saisai Shao > Attachments: SortBasedShuffleRead.pdf, SortBasedShuffleReader on > Spark 2.x.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test > Report.pdf > > > Currently Spark has already integrated sort-based shuffle write, which > greatly improve the IO performance and reduce the memory consumption when > reducer number is very large. But for the reducer side, it still adopts the > implementation of hash-based shuffle reader, which neglects the ordering > attributes of map output data in some situations. > Here we propose a MR style sort-merge like shuffle reader for sort-based > shuffle to better improve the performance of sort-based shuffle. > Working in progress code and performance test report will be posted later > when some unit test bugs are fixed. > Any comments would be greatly appreciated. > Thanks a lot. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16265035#comment-16265035 ] Saisai Shao commented on SPARK-2926: So your saying of 12x-30x boosting is only referring to reduce stage? of course this solution can boost the reduce stage, but it will also increase the time of map stage, so we'd better to use job time to evaluate. > Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle > -- > > Key: SPARK-2926 > URL: https://issues.apache.org/jira/browse/SPARK-2926 > Project: Spark > Issue Type: Improvement > Components: Shuffle >Affects Versions: 1.1.0 >Reporter: Saisai Shao >Assignee: Saisai Shao > Attachments: SortBasedShuffleRead.pdf, SortBasedShuffleReader on > Spark 2.x.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test > Report.pdf > > > Currently Spark has already integrated sort-based shuffle write, which > greatly improve the IO performance and reduce the memory consumption when > reducer number is very large. But for the reducer side, it still adopts the > implementation of hash-based shuffle reader, which neglects the ordering > attributes of map output data in some situations. > Here we propose a MR style sort-merge like shuffle reader for sort-based > shuffle to better improve the performance of sort-based shuffle. > Working in progress code and performance test report will be posted later > when some unit test bugs are fixed. > Any comments would be greatly appreciated. > Thanks a lot. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[
https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16265030#comment-16265030
]
Li Yuanjian commented on SPARK-2926:
[~jerryshao], thanks a lot for your advise and reply.
{quote}
would you please use spark-perf's micro benchmark
(https://github.com/databricks/spark-perf) to verify again with same workload
as mentioned in original test report?
{quote}
Sure, I'll verify this again ASAP.
{quote}
Theoretically this solution cannot get 12x-30x boosting according to my test
{quote}
Firstly I also had question on this, I attached all the screenshot in the pdf.
The 12x boosting happened in both scenario of reducer task number is 1 and 100.
The duration of this stage reduce from 2min to 9s(13x) while reducer task
number is 1 and reduce from 20min to 1.4min while the number is 100. The 30x
boosting happened after I add more data pressure for reducer task.
{quote}
Can you please explain the key difference and the reason of such boosting?
{quote}
I think the key difference mainly comes from this 2 points:
1. Like saisai said, BlockStoreShuffleReader use `ExternalSorter` deal with the
reduce work, each record should do the compare work, while SortShuffleReader is
more cpu friendly, it collect all shuffle map result(both data in memory and
data spilled to disk) and sort them by merging sort(each partition has been
sorted in map side).
2. The obvious cut down of peak memory used in reduce task, this will save gc
time during sorting.
> Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
> --
>
> Key: SPARK-2926
> URL: https://issues.apache.org/jira/browse/SPARK-2926
> Project: Spark
> Issue Type: Improvement
> Components: Shuffle
>Affects Versions: 1.1.0
>Reporter: Saisai Shao
>Assignee: Saisai Shao
> Attachments: SortBasedShuffleRead.pdf, SortBasedShuffleReader on
> Spark 2.x.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test
> Report.pdf
>
>
> Currently Spark has already integrated sort-based shuffle write, which
> greatly improve the IO performance and reduce the memory consumption when
> reducer number is very large. But for the reducer side, it still adopts the
> implementation of hash-based shuffle reader, which neglects the ordering
> attributes of map output data in some situations.
> Here we propose a MR style sort-merge like shuffle reader for sort-based
> shuffle to better improve the performance of sort-based shuffle.
> Working in progress code and performance test report will be posted later
> when some unit test bugs are fixed.
> Any comments would be greatly appreciated.
> Thanks a lot.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16264904#comment-16264904 ] Saisai Shao commented on SPARK-2926: [~XuanYuan], would you please use spark-perf's micro benchmark (https://github.com/databricks/spark-perf) to verify again with same workload as mentioned in original test report? That would be more comparable. Theoretically this solution cannot get 12x-30x boosting according to my test, because this solution don't actually reduce the computation in logic, just moving part of comparison from reduce to map, which potentially reduces some cpu cycling and improves cache hit. Can you please explain the key difference and the reason of such boosting? Thanks! > Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle > -- > > Key: SPARK-2926 > URL: https://issues.apache.org/jira/browse/SPARK-2926 > Project: Spark > Issue Type: Improvement > Components: Shuffle >Affects Versions: 1.1.0 >Reporter: Saisai Shao >Assignee: Saisai Shao > Attachments: SortBasedShuffleRead.pdf, SortBasedShuffleReader on > Spark 2.x.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test > Report.pdf > > > Currently Spark has already integrated sort-based shuffle write, which > greatly improve the IO performance and reduce the memory consumption when > reducer number is very large. But for the reducer side, it still adopts the > implementation of hash-based shuffle reader, which neglects the ordering > attributes of map output data in some situations. > Here we propose a MR style sort-merge like shuffle reader for sort-based > shuffle to better improve the performance of sort-based shuffle. > Working in progress code and performance test report will be posted later > when some unit test bugs are fixed. > Any comments would be greatly appreciated. > Thanks a lot. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16251457#comment-16251457 ] Li Yuanjian commented on SPARK-2926: I just giving a preview PR above, I'll collect more suggestions about this and maybe raise a SPIP vote later. > Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle > -- > > Key: SPARK-2926 > URL: https://issues.apache.org/jira/browse/SPARK-2926 > Project: Spark > Issue Type: Improvement > Components: Shuffle >Affects Versions: 1.1.0 >Reporter: Saisai Shao >Assignee: Saisai Shao > Attachments: SortBasedShuffleRead.pdf, SortBasedShuffleReader on > Spark 2.x.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test > Report.pdf > > > Currently Spark has already integrated sort-based shuffle write, which > greatly improve the IO performance and reduce the memory consumption when > reducer number is very large. But for the reducer side, it still adopts the > implementation of hash-based shuffle reader, which neglects the ordering > attributes of map output data in some situations. > Here we propose a MR style sort-merge like shuffle reader for sort-based > shuffle to better improve the performance of sort-based shuffle. > Working in progress code and performance test report will be posted later > when some unit test bugs are fixed. > Any comments would be greatly appreciated. > Thanks a lot. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16251452#comment-16251452 ] Apache Spark commented on SPARK-2926: - User 'xuanyuanking' has created a pull request for this issue: https://github.com/apache/spark/pull/19745 > Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle > -- > > Key: SPARK-2926 > URL: https://issues.apache.org/jira/browse/SPARK-2926 > Project: Spark > Issue Type: Improvement > Components: Shuffle >Affects Versions: 1.1.0 >Reporter: Saisai Shao >Assignee: Saisai Shao > Attachments: SortBasedShuffleRead.pdf, SortBasedShuffleReader on > Spark 2.x.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test > Report.pdf > > > Currently Spark has already integrated sort-based shuffle write, which > greatly improve the IO performance and reduce the memory consumption when > reducer number is very large. But for the reducer side, it still adopts the > implementation of hash-based shuffle reader, which neglects the ordering > attributes of map output data in some situations. > Here we propose a MR style sort-merge like shuffle reader for sort-based > shuffle to better improve the performance of sort-based shuffle. > Working in progress code and performance test report will be posted later > when some unit test bugs are fixed. > Any comments would be greatly appreciated. > Thanks a lot. -- This message was sent by Atlassian JIRA (v6.4.14#64029) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[
https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16251398#comment-16251398
]
Li Yuanjian commented on SPARK-2926:
During our work of migrating some old Hadoop job to Spark, I noticed this JIRA
and the code based on spark 1.x.
I re-implemented the old PR based on Spark 2.1 and current master branch. After
produced some scenario and ran some benchmark tests, I found that this shuffle
mode can bring {color:red}12x~30x boosting in task duration and reduce peak
execution memory to 1/12 ~ 1/50{color} vs current master version, especially
the memory reducing, in this shuffle mode Spark can support more data size in
less memory usage. The detail doc attached in this jira named
"SortShuffleReader on Spark 2.x".
I know that DataSet API will have better optimization and performance, but RDD
API may still useful for flexible control and old Spark/Hadoop jobs. For the
better performance in ordering cases and more cost-effective memory usage,
maybe this PR is still worth to merge in to master.
I'll sort out current code base and give a PR soon. Any comments and trying out
would be greatly appreciated.
> Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
> --
>
> Key: SPARK-2926
> URL: https://issues.apache.org/jira/browse/SPARK-2926
> Project: Spark
> Issue Type: Improvement
> Components: Shuffle
>Affects Versions: 1.1.0
>Reporter: Saisai Shao
>Assignee: Saisai Shao
> Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test
> Report(contd).pdf, Spark Shuffle Test Report.pdf
>
>
> Currently Spark has already integrated sort-based shuffle write, which
> greatly improve the IO performance and reduce the memory consumption when
> reducer number is very large. But for the reducer side, it still adopts the
> implementation of hash-based shuffle reader, which neglects the ordering
> attributes of map output data in some situations.
> Here we propose a MR style sort-merge like shuffle reader for sort-based
> shuffle to better improve the performance of sort-based shuffle.
> Working in progress code and performance test report will be posted later
> when some unit test bugs are fixed.
> Any comments would be greatly appreciated.
> Thanks a lot.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14379532#comment-14379532 ] Saisai Shao commented on SPARK-2926: Hi [~DoingDone9], would you please give some detailed information about your test environment, like cluster size, hardware configurations, as well as Spark configurations. Also would you please offer each stage's running time as compared to total running time. Thanks a lot. As for my local environment with small 1 master + 4 slaves, I tested with my patch rebased to the latest master, as compared to the master branch, the result shows that the performance of sortByKey with my patch is still faster than the master branch at about (15% to 20 %). I think there's a possibility that different hardware configurations may shift the hardware bottleneck and result in different results, I will investigate more, it would be very helpful if you could offer some more detailed information. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Assignee: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[
https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14375323#comment-14375323
]
DoingDone9 commented on SPARK-2926:
---
hi, i test sortByKey with spark-perf(https://github.com/databricks/spark-perf),
but i have a result like that :
spark1.3 :
{time:452.453},{time:457.929},{time:452.295}
with your pr
{time:471.215},{time:460.59},{time:463.795}
could you tell me something taht i did incorretly. Thank you.
Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
--
Key: SPARK-2926
URL: https://issues.apache.org/jira/browse/SPARK-2926
Project: Spark
Issue Type: Improvement
Components: Shuffle
Affects Versions: 1.1.0
Reporter: Saisai Shao
Assignee: Saisai Shao
Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test
Report(contd).pdf, Spark Shuffle Test Report.pdf
Currently Spark has already integrated sort-based shuffle write, which
greatly improve the IO performance and reduce the memory consumption when
reducer number is very large. But for the reducer side, it still adopts the
implementation of hash-based shuffle reader, which neglects the ordering
attributes of map output data in some situations.
Here we propose a MR style sort-merge like shuffle reader for sort-based
shuffle to better improve the performance of sort-based shuffle.
Working in progress code and performance test report will be posted later
when some unit test bugs are fixed.
Any comments would be greatly appreciated.
Thanks a lot.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
-
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14285272#comment-14285272 ] Saisai Shao commented on SPARK-2926: Hi [~manojsamel], we're still waiting for any maintainer who could review this, thanks for your attention. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Assignee: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14284300#comment-14284300 ] Manoj Samel commented on SPARK-2926: Which release will have this change available ? Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Assignee: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14223925#comment-14223925 ] Apache Spark commented on SPARK-2926: - User 'jerryshao' has created a pull request for this issue: https://github.com/apache/spark/pull/3438 Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Assignee: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14179631#comment-14179631 ] Sandy Ryza commented on SPARK-2926: --- [~rxin] did you ever get a chance to try this out? Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14138547#comment-14138547 ] Saisai Shao commented on SPARK-2926: Looking forward to your feedback :). Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14133694#comment-14133694 ] Saisai Shao commented on SPARK-2926: Hey [~rxin], here is the branch rebased on your code (https://github.com/jerryshao/apache-spark/tree/sort-shuffle-read-new-netty), mind taking a look at it? Thanks a lot. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14132598#comment-14132598 ] Saisai Shao commented on SPARK-2926: Ok, I will take a try and let you know then it is ready. Thanks a lot. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14131314#comment-14131314 ] Saisai Shao commented on SPARK-2926: Hi Reynold, thanks a lot for your watching this, here is the branch (https://github.com/jerryshao/apache-spark/tree/sort-based-shuffle-read), though code is not rebase to the latest master branch. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14131823#comment-14131823 ] Reynold Xin commented on SPARK-2926: Do you mind creating a separate branch that's based on https://github.com/rxin/spark/tree/netty-blockTransferService ? Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report(contd).pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14122292#comment-14122292 ] Saisai Shao commented on SPARK-2926: Hi Matei, sorry for late response, I will test more scenarios with your notes, also factor out to see if some codes can be shared with ExternalSorter. Thanks a lot. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14111631#comment-14111631 ] Matei Zaharia commented on SPARK-2926: -- I see, thanks for posting the benchmarks. This does seem like it's worth investigating further. Can you also run some tests with other aggregation factors? Also, a few notes on the configuration: * There's a big change in behavior when you go above 200 reduce tasks because ExternalSorter does the same thing as hash-based shuffle if the # of reduce tasks is below 200. * When doing these kind of tests, set spark.kryo.referenceTracking = false, otherwise serialization will be a major CPU cost. * We need to try other types of keys as well, e.g. integers or longer strings. As I said the cost to compare elements will depend on their datatype. Finally, it would be great if this proposal reused parts of ExternalSorter or otherwise shared code for it. It looks like it's not a clear win in all cases, but maybe for example we can use it in sortByKey, and later in groupBy / join when we update those to deal with it. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.2#6252) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14096688#comment-14096688 ] Saisai Shao commented on SPARK-2926: I think this prototype can easily offer the functionality SPARK-2978 needed. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf, Spark Shuffle Test Report.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.2#6252) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14091684#comment-14091684 ] Saisai Shao commented on SPARK-2926: Hi Sandy, Thanks a lot for your comments, basic idea is the same as MapReduce, but the implementation is a little different to be compatible with Spark's code path. Code may has some common parts with map-side merge, I think currently we should verify the pros and cons of this method, then we can do the refactoring to make it better. For operation like groupByKey, I'm not sure the performance of by-key sorting is better than Aggregator way or not, currently I still use Aggregator in reduce side. We need to do some performance tests to verify this by-key sorting's necessity. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.2#6252) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14091685#comment-14091685 ] Matei Zaharia commented on SPARK-2926: -- Hey Saisai, a couple of questions about this: - Doesn't the ExternalAppendOnlyMap used on the reduce side already do a merge-sort? You won't do much better than that unless you assume that an Ordering is given for the key, which isn't actually something we receive at the ShuffleWriter from our APIs so far (though we have the opportunity to pass it through in Scala). - What kind of data did you test key sorting on the map side with? The cost of comparisons will depend a lot on the data type, but will be much higher for things like strings or tuples. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.2#6252) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14091753#comment-14091753 ] Saisai Shao commented on SPARK-2926: Hi Matei, thanks a lot for your comments. The original point of this proposal is to directly merge the data without re-sorting (when spilling using EAOM) if data is by-key sorted (with keyOrdering or hashcode ordering) in map outputs. If there is no Ordering available or needed like groupByKey, current design thinking is still using EAOM to do aggregation. I use SparkPerf sort-by-key workload to test the current shuffle implementations: 1. sort shuffle write with hash shuffle read (current sort-based shuffle implementation). 2. sort shuffle write and sort merge shuffle read (my prototype). Test data type is String, key and value length is 10, and record number is 2G, data is stored in HDFS. My rough test shows that my prototype may be slower in shuffle write (1.18x slower) because of another key comparison, but 2.6x faster than HashShuffleReader in reduce side. I have to admit that only sort-by-key cannot well illustrate the necessity of this proposal, also the method is better for sortByKey scenario. I will continue to do some other workload tests to see if this method is really necessary or not. I will post my test result later. Thanks again for your comments. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.2#6252) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
[jira] [Commented] (SPARK-2926) Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle
[ https://issues.apache.org/jira/browse/SPARK-2926?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14091489#comment-14091489 ] Sandy Ryza commented on SPARK-2926: --- Hi Saisai, This seems like a very useful addition. If my understanding is correct, the reduce-side merge procedure seems like a fairly straightforward port of the approach used in MapReduce. A couple considerations: * Is there code we can share with the map-side merge? * It may make sense to use by-key sorting in some situations even when the operation doesn't require a key ordering. For example, if we're doing a groupByKey and are going to end up spilling most of the data on the reduce side anyway, holding objects in Aggregator's in-memory hashmap is pure overhead. Add MR-style (merge-sort) SortShuffleReader for sort-based shuffle -- Key: SPARK-2926 URL: https://issues.apache.org/jira/browse/SPARK-2926 Project: Spark Issue Type: Improvement Components: Shuffle Affects Versions: 1.1.0 Reporter: Saisai Shao Attachments: SortBasedShuffleRead.pdf Currently Spark has already integrated sort-based shuffle write, which greatly improve the IO performance and reduce the memory consumption when reducer number is very large. But for the reducer side, it still adopts the implementation of hash-based shuffle reader, which neglects the ordering attributes of map output data in some situations. Here we propose a MR style sort-merge like shuffle reader for sort-based shuffle to better improve the performance of sort-based shuffle. Working in progress code and performance test report will be posted later when some unit test bugs are fixed. Any comments would be greatly appreciated. Thanks a lot. -- This message was sent by Atlassian JIRA (v6.2#6252) - To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org
