Github user neoremind commented on the issue:
https://github.com/apache/spark/pull/18964
@zsxwing I did try to create a performance test against spark rpc, the test
result can be found
[here](https://github.com/neoremind/kraps-rpc#4-performance-test), note that I
created the project for studying purpose and the code is based on 2.1.0. But as
you said, the performance would not be dropped as client not using `SO_RCVBUF`
and `SO_SNDBUF` set in `SparkConf`.
For example, I use the scenario of concurrent calls 10, total calls 100k,
keep all things as default, the QPS would be around 11k. When I set `SO_RCVBUF`
and ` SO_SNDBUF` to extremely small number like 100 the performance is
affected tremendously. If they are set to a large number like 128k, the results
won't be boosted by whether clients set the corresponding `SO_RCVBUF` and
`SO_SNDBUF` value or not.
I admit that the update is trivial but from users' perspective, if
`spark.{module}.io.sendBuffer` and `spark.{module}.io.sendBuffer` are exposed
outside and could be set, and they only works on server side, I think it is a
little bit not consistent, so I raise the PR to try to make it work on both
server and client side, just to make them consistent.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
