Re: [Pound Mailing List] Threads vs connections

"D. Hampton Finger" <[email protected]> Wed, 9 Dec 2015 09:13:28 -0600 (CST)
Newsgroups gmane.comp.web.pound.general
Message-ID <[email protected]>
I have sought guidance on this subject, read a lot, and ended up guessing 
and monitoring my setup for how those settings hit my performance.  We 
take around 1.3 million submissions per week on average and have pound 
running on an active/passive setup with keepalived.  Pound is setup for 
600 max threads.

Traffic as monitored by Munin over the past year using a custom plugin 
reading "poundctl -H -c /var/lib/pound/pound.cfg":

Sessions Max: 5249.27
Sessions Average: 984.91

Threads Max: 600
Threads Running: 22.22
Threads Sleeping: 580.71

I might think about changing that number down, but it has shown in the 
past that if there are problems with my setup (DB latency, authentication 
machinery stalling) that the thread count will climb and trigger a nagios 
alert faster than any other metric on my stack alerting me to impending 
issues.  I did find that keepalives on the back-end servers (apache2) did 
massively increase the thread use on the pound systems by an order of 
magnitude or two.  I have them off, and it has benefits to resource use 
for my application.

so TL;DR:  Guess, monitor, adujust, monitor, adjust etc.

Hampton

On Wed, 9 Dec 2015, John Hayward wrote:

> 
> We currently have pound configured for 100 threads.
> 
> 
> We typically have at least 70 to 90+ sessions shown by poundctl.
> 
> 
> Is there only one session per thread or can a thread support multiple
> sessions?
> 
> 
> If multiple sessions then is there a best practice on max number of sessions
> per thread one should allow?
> 
> 
> johnh... 
> 
> 
>

--
To unsubscribe send an email with subject unsubscribe to [email protected]
Please contact [email protected] for questions.