Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[SPARK-5208][DOC] Add more documentation to Netty-based configs #4012

Closed
wants to merge 1 commit into from

Conversation

sarutak
Copy link
Member

@sarutak sarutak commented Jan 13, 2015

#3713 added some documentation about Netty-based configs but I think we need more. I think following configs can be useful for performance tuning.

  • spark.shuffle.io.mode
  • spark.shuffle.io.backLog
  • spark.shuffle.io.receiveBuffer
  • spark.shuffle.io.sendBuffer

@SparkQA
Copy link

SparkQA commented Jan 13, 2015

Test build #25437 has finished for PR 4012 at commit a6e2aa7.

  • This patch passes all tests.
  • This patch merges cleanly.
  • This patch adds no public classes.

@sarutak sarutak changed the title [SPARK-5208][DOCS] Add more documentation to Netty-based configs [SPARK-5208][DOC] Add more documentation to Netty-based configs Jan 13, 2015
@pwendell
Copy link
Contributor

@sarutak when we added the netty shuffle we actually decided not to expose these in order to keep the overall # of configurations manageable. We couldn't think of a user scenario where these would make a large difference (correct me if that is wrong @aarondav).

Did you have a specific use case in mind, or was this mostly for completeness reasons?

@aarondav
Copy link
Contributor

I agree with Patrick -- it's a very case-by-case basis, but we decided not to document these:

  • spark.shuffle.io.mode: We could not see a difference between NIO and EPOLL from a performance perspective, and EPOLL is less stable, so we did not want to suggest to use it.
  • spark.shuffle.io.backLog: This is more suitable for a serving layer than a shuffle layer, I can't think why someone would want to specify this.
  • spark.shuffle.io.receiveBuffer/sendBuffer: These are very low-level tuning parameters that are likely not to achieve a noticeable performance benefit, so documenting them might just lead people to think they're more important than they are.

@pwendell
Copy link
Contributor

pwendell commented Feb 2, 2015

Okay @sarutak can you close this issue then? Looks like we intentionally left these out for now

@sarutak
Copy link
Member Author

sarutak commented Feb 2, 2015

O.K, I'll close.

@sarutak sarutak closed this Feb 2, 2015
@sarutak sarutak deleted the more-description-for-netty branch April 11, 2015 05:24
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants