[
https://issues.apache.org/jira/browse/CASSANDRA-21645?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Francisco Guerrero updated CASSANDRA-21645:
-------------------------------------------
Status: Needs Committer (was: Patch Available)
> Use MPSC queue in native transport flusher
> ------------------------------------------
>
> Key: CASSANDRA-21645
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21645
> Project: Apache Cassandra
> Issue Type: Improvement
> Components: Messaging/Client
> Reporter: Dmitry Konstantinov
> Assignee: Dmitry Konstantinov
> Priority: Normal
> Fix For: 6.0.x, 7.0
>
> Attachments: mpsc_flusher_after_cpu.html, mpsc_flusher_before_cpu.html
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> org.apache.cassandra.transport.Flusher#queued is consumed by single thread -
> Netty even loop thread responsible for the correspondent channel handling, so
> we can replace ConcurrentLinkedQueue with more specialized version such as
> org.jctools.queues.MpscUnboundedArrayQueue. It also remove an allocation
> present for ConcurrentLinkedQueue.
> CPU usage in flamegraph ("Flusher\.enqueue" pattern), before: 1.17%, after:
> 0.46%
> [^mpsc_flusher_before_cpu.html]
> [^mpsc_flusher_after_cpu.html]
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
