ivandika3 commented on PR #1455: URL: https://github.com/apache/ratis/pull/1455#issuecomment-4523645142
Thanks @szetszwo for checking this out. Did some benchmark for the first iteration where we batch by time and size and there are some improvements, but need to reduce the batch time to 50 microseconds. However, even then, the pure follower reads performance still cannot beat the pure leader read performance. I will try to benchmark the natural batching method. However, currently I'm still preoccupied with other tasks so it won't be done so soon. If even with natural batching it doesn't work, we might need to go one level deeper to reduce ReadIndex latency using things like IO_URING ([RATIS-2545](https://issues.apache.org/jira/browse/RATIS-2545)) or even changing the architecture to use SEDA similar to Sofa-Jraft. However, please let me know if there are some bottlenecks that I missed. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
