Github user uce commented on the pull request:
https://github.com/apache/flink/pull/1093#issuecomment-138237365
Impressive! :-) Thanks for the write up. I think I've learned something
about the JVM by reading your PR description. ;)
With some minor refactorings in the network stack, we could get rid of a
memory copy before putting off-heap segments to the wire. Instead of returning
`Buffer` instances we would return `ByteBuf` instances (Netty's recyclable
Buffer API), which could then be handed to Netty directly. Currently, we copy
from our `Buffer `s to Netty's `ByteBuf`s. I would not expect a huge
performance improvement, but I think there are other good reasons to wrap
memory in `ByteBuf`s. This is not urgent, but something I wanted to try out for
some time now, maybe I'll try something on a free weekend.
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
