[Freeswitch-users] RedHat 6.X Performance

[Michael Giagnocavo]

I found someone with more info on scheduling problems, especially with sleep/wait, of which FS does a lot. There's a parameter called sched_migration_cost which might be worth investigating.

http://pastebin.freeswitch.com/19903

On a benchmark for sleeps, he says RHEL6 is taking up to 45% more time. With the huge number of threads and sleeps FS does, maybe those little differences add up:
http://unix.stackexchange.com/questions/37391/high-cpu-usage-with-cfs

Separately, to everyone that hasn't rebooted their server in the last few months, you may still have the leap-second set which will cause lots of CPU spinning due to a kernel bug. Running date -s "`date`" should fix that. I am pretty sure that freed up at least one core on one of our boxes that was still spinning since Jul 1.

We've now upgraded some systems to CentOS 6.3 (with all post release updates) and left others at 6.2. Over the next week we'll see if they seem to be much different. At the moment, the general CPU usage seems to be similar; we'll have to pound some dialer on it and see what happens.

-Michael


-----Original Message-----
From: freeswitch-users-bounces at lists.freeswitch.org [mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf Of Kristian Kielhofner
Sent: Monday, September 03, 2012 12:42 PM
To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] RedHat 6.X Performance

Anymore info on specific traffic patterns that cause this?  I've been beating up FS on 6.3 for several days now and haven't seen anything unusual yet...

On Thu, Aug 30, 2012 at 11:45 PM, Ken Rice <krice at freeswitch.org> wrote:
> I'm running probably one of the most stripped down FreeSWITCH configs 
> you can run ... Sofia only, bypass media, with a custom C routing 
> module that uses libpq directly...
>
> the problem happens at somewhere around 50 to 100 CPS, system % goes 
> thru the roof, and its not IO.. loglevel 0... etc... now its also 
> worth mentioning that on 6.2 the number of context switches for 
> similar amounts of calls on the same physical hardware seems to be 
> double vs something like
> cent5 or deb6....
>
> we're currently trying to get people to submit reports from oprofile 
> so we can isolate the issue... if it were just me, I would chalk it up 
> to something specific on the installation, but , the number of 
> reporters with similar observations is something to take notice of...
>
>

--
Kristian Kielhofner

_________________________________________________________________________
Professional FreeSWITCH Consulting Services:
consulting at freeswitch.org
http://www.freeswitchsolutions.com

 

Official FreeSWITCH Sites
http://www.freeswitch.org
http://wiki.freeswitch.org
http://www.cluecon.com

FreeSWITCH-users mailing list
FreeSWITCH-users at lists.freeswitch.org
http://lists.freeswitch.org/mailman/listinfo/freeswitch-users
UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users
http://www.freeswitch.org