[
https://issues.apache.org/jira/browse/TS-1267?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13289968#comment-13289968
]
Zhao Yongming commented on TS-1267:
-----------------------------------
in cluster, the cluster connection is channel reusing bi-direction connection,
very different from the http connections, it may handle 6k+ virtual channel in
the single connections, so in most case, the connections may not be the core
problem.
we need to change the connections numbers because we find that the cpu usage of
one connections(it will bind to one thread, then bind to one cpu), is the
bottleneck when you running a heavy site, ie: you have 4 boxes and runing 10G
traffic, before we change that, it will kill 3 cpus on each host, can not take
benefit from other cpus.
in our discuss, we are considering implement the thread<->thread connections
between each box, or 1+ connections maybe, does that sounds reasonable?
> Should we have cache cluster connection pools per EThread ?
> -----------------------------------------------------------
>
> Key: TS-1267
> URL: https://issues.apache.org/jira/browse/TS-1267
> Project: Traffic Server
> Issue Type: New Feature
> Components: Clustering
> Reporter: Leif Hedstrom
> Fix For: 3.3.0
>
>
> Once TS-1222 lands, we might want to look into the possibility of having
> per-EThread cluster connection pools. This assures that the client VC and the
> cluster VC are on the same thread. Similar to how we optimize origin VC
> connections.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
