[
https://issues.apache.org/jira/browse/TS-2994?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14534550#comment-14534550
]
Leif Hedstrom commented on TS-2994:
-----------------------------------
Yeah, I think we should do this for both of our accept cases:
1) Accept on ET_NET
2) Dedicated accept thread (when m > 1 of course).
#2 becomes even more important if/when we go to a model of one accept thread
per NUMA node. I read the HTTPD dev@ mailing list thread, and they are seeing
some amazing performance improvements. From my own experience, since we epoll()
on the FD when we accept in ET_NET, we sometimes (often?) seem to do a lot of
expensive accept() for no good reason.
There's an Intel article on the subject as well:
http://www.intel.ie/content/dam/www/public/us/en/documents/white-papers/scaling-apache-server-performance-paper.pdf
> Make use of SO_REUSEPORT socket option
> --------------------------------------
>
> Key: TS-2994
> URL: https://issues.apache.org/jira/browse/TS-2994
> Project: Traffic Server
> Issue Type: New Feature
> Components: Core
> Reporter: Yunkai Zhang
> Assignee: Phil Sorber
> Fix For: 6.0.0
>
>
> "One of the features merged in the 3.9 development cycle was TCP and UDP
> support for the SO_REUSEPORT socket option; that support was implemented in a
> series of patches by Tom Herbert. The new socket option allows multiple
> sockets on the same host to bind to the same port, and is intended to improve
> the performance of multithreaded network server applications running on top
> of multicore systems." --[The SO_REUSEPORT socket
> option|http://lwn.net/Articles/542629/]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
