Performance improvement, moving write and flush out of enframed and i… #108
Conversation
…nto framed write itself
|
Sweet! Any measurement of by how much this improves throughput? |
|
Nothing formal; however, the benchmark tests for our application's use of remotely showed significant improvement. An exercise of 60 concurrent RPC calls (where the payload size varied between ~30 and 1024 bytes) took 1/10th the time to execute. And a more complete integration test suite of 125 tests that talk to two RPC servers (one handling persistence and one executing downstream messaging to clients) dropped from nearly 3 minutes to less than 30 seconds. |
|
What is the status of this? :) Btw @sbuzzard I have an open PR that bumps netty to 4.1.0 |
|
@ahjohannessen - Not sure. We have been working with our own fork and I had forgotten to follow up. This change has made a world of difference in performance with our applications, but I'd definitely like to try with netty 4.1.0. I'm guessing it hasn't been merged since the travis builds are failing, but I don't think those failures are (at least directly) related to this change ... |
|
@sbuzzard Ok. You are correct with travis failures, they have been happening for a long time now. |
|
@sbuzzard @ahjohannessen sorry for the delay on this - i had not noticed the PR (subsequently fixed my github notification settings). This looks like another great change |
…nto framed write itself