[
https://issues.apache.org/jira/browse/TS-1107?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Bryan Call closed TS-1107.
--------------------------
Resolution: Won't Fix
I don't think this is really needed. People should really know how much
traffic they are going to get on the server upfront. httpd depended on scaling
the number of processes in the past and it never really worked well.
It would be difficult to know when to create new threads and when you need them
the most, their proxy allocators would be empty.
> dynamically scale the number of net threads
> -------------------------------------------
>
> Key: TS-1107
> URL: https://issues.apache.org/jira/browse/TS-1107
> Project: Traffic Server
> Issue Type: New Feature
> Components: Core, Performance
> Reporter: James Peach
> Priority: Minor
> Fix For: 6.0.0
>
>
> The number of net threads is calculated once at startup, but we ought to
> consider dynamically scaling the number of threads a runtime based on load.
> zwoop: right, that's what I meant (keep a counter of how many times epoll had
> no events, and treat that as an "idle" thread)
> zwoop: probably a multiplier of some setting (thread_idle_seconds or some
> such)
> zwoop: this would be a cool feature, if you have the time for it ;)
> zwoop: can keep the original calculations / settings as the "upper" limit I
> think
> zwoop: (the calculations can also easily be configured in records.config, so
> that people can modify that upper limit)
> zwoop: an ideal solution (backward compatible) would be to just add one new
> setting, <reap_idle_threads_seconds> or some such, default to 0 (off). With
> it not set, our normal logic applies. With it set, your stuff takes effect,
> but cap it at whatever the other settings are.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
