[
https://issues.apache.org/jira/browse/FLINK-9913?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16580829#comment-16580829
]
ASF GitHub Bot commented on FLINK-9913:
---------------------------------------
zhijiangW edited a comment on issue #6417: [FLINK-9913][runtime] Improve output
serialization only once in RecordWriter
URL: https://github.com/apache/flink/pull/6417#issuecomment-413127511
For benchmark, I also create a new job with only source and map vertex, and
the source broadcast Long type value to all the maps. I execute the same job
for setting the parallelism 1:100 and 1:200 separately, and the throughput
increases 13%, 15% separately in the following:
```
parallelism | throughput (before) | throughput (now)
1:100 | 70.760 ± 10.557 ops/ms | 83.480 ± 1.967 ops/ms
1:200 | 37.756 ± 1.170 ops/ms | 43.316 ± 2.176 ops/ms
```
In theory, we only reduce the number of serialization times and do not
introduce any time-cost operations. I will further verify the
`StreamNetworkThroughputBenchmarkExecutor` as you mentioned and show your
results after done. :)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> Improve output serialization only once in RecordWriter
> ------------------------------------------------------
>
> Key: FLINK-9913
> URL: https://issues.apache.org/jira/browse/FLINK-9913
> Project: Flink
> Issue Type: Improvement
> Components: Network
> Affects Versions: 1.6.0
> Reporter: zhijiang
> Assignee: zhijiang
> Priority: Minor
> Labels: pull-request-available
> Fix For: 1.7.0
>
>
> Currently the {{RecordWriter}} emits output into multi channels via
> {{ChannelSelector}} or broadcasts output to all channels directly. Each
> channel has a separate {{RecordSerializer}} for serializing outputs, that
> means the output will be serialized as many times as the number of selected
> channels.
> As we know, data serialization is a high cost operation, so we can get good
> benefits by improving the serialization only once.
> I would suggest the following changes for realizing it.
> # Only one {{RecordSerializer}} is created in {{RecordWriter}} for all the
> channels.
> # The output is serialized into the intermediate data buffer only once for
> different channels.
> # The intermediate serialization results are copied into different
> {{BufferBuilder}}s for different channels.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
