[Freeswitch-users] Freeswitch Performance

Andre andretodd at verizon.net
Fri Nov 22 17:25:28 MSK 2013


After all day of testing yesterday (max 100 cps) it appears that odbc for
windows won't work. It must be a bug because if I set the core db, profiles
etc. to use odbc to MSSQL it crashes after 1 or 2 minutes of cps of 50. If I
remove that odbc connection and use the built in databases it works like a
champ.  Yes I do still see a small memory leak still so that will need to be
discovered and fixed.

 

I think it's what you said about too many transitions and it can't keep up.

Andre

 

From: freeswitch-users-bounces at lists.freeswitch.org
[mailto:freeswitch-users-bounces at lists.freeswitch.org] On Behalf Of Steven
Ayre
Sent: Friday, November 22, 2013 9:13 AM
To: FreeSWITCH Users Help
Subject: Re: [Freeswitch-users] Freeswitch Performance

 

It didn't explictly say 'pause'... what i noticed grep-ing the log for the
uuid of a call's that had hung was that it was going through a sequence of
log messages that are usually <1ms apart and instead were ~30s apart.

 

-Steve

 

On 20 November 2013 21:56, Andre Demattia <andretodd at verizon.net
<mailto:andretodd at verizon.net> > wrote:

I had debug turned off but I'll enable it to see if anything interesting
happens.
I didn't see anything that said pause.

I'll also turn off odbc for the core db and use the default one to see if
that helps. Un testing it didn't help but I'll try it in again. 

  _____  

From: Steven Ayre <mailto:steveayre at gmail.com> 
Sent: 11/20/2013 4:36 PM
To: FreeSWITCH Users Help <mailto:freeswitch-users at lists.freeswitch.org> 


Subject: Re: [Freeswitch-users] Freeswitch Performance

> 


> show nothing other than FreeSWITCH was receiving the new messages but no
> bridge or cancel actions were taking place


Do you have your logs at debug level and are there any log messages that
indicate a call reaching a point then pausing?

I saw something recently similar under load, and it turned out to be the
limit subsystem was waiting on DB queries that were severely backing up
because there were too many commits happening per second and InnoDB was
configured to flush to disk on every transaction commit which can only
happen ~200 times per second. (Switching to the memory engine fixed the
error although isn't transactional).

-Steve

-Steve


On 20 November 2013 20:38, Andre <andretodd at verizon.net
<mailto:andretodd at verizon.net> > wrote:

> I tested again and when I use 25 cps for 2 hours everything worked fine.
> When I sent 50 cps I noticed FreeSWITCH worked for about 5 minutes then I
> saw the memory jump from 122 mb to to 1.6 gb in seconds. I think
FreeSWITCH
> crashed but the logs show nothing other than FreeSWITCH was receiving the
> new messages but no bridge or cancel actions were taking place.
>
>
>
> How do I tell what happened that made freeswitch just stop working? The
> GUI didn't show anything like a throw exception and when I type version or
> sofia status in the FreeSWITCH gui it doesn't respond.
>
>
>
> Andre
>
>
>

> *From:* freeswitch-users-bounces at lists.freeswitch.org
<mailto:freeswitch-users-bounces at lists.freeswitch.org>  [mailto:
> freeswitch-users-bounces at lists.freeswitch.org
<mailto:freeswitch-users-bounces at lists.freeswitch.org> ] *On Behalf Of
*Anthony
> Minessale
> *Sent:* Wednesday, November 20, 2013 12:33 AM
> *To:* Freeswitch-users
>
> *Subject:* Re: [Freeswitch-users] Freeswitch Performance


>
>
>
> Get HEAD of stable branch or HEAD of master for latest build files.
>
> There is no promise windows can do 500cps the industry standard is 50 and
> you'll have probably more luck on linux but there is only on way to fimd
> out.
>
> On Nov 19, 2013 11:27 PM, "I put the Who? in Mishehu" <
> mishehu at freeswitch.org <mailto:mishehu at freeswitch.org> > wrote:
>
> The amusing thing about the recurring load test threads that pop up every
> now and then is that the most important tips are already documented on the
> wiki in one single page.  Beyond that, you probably spend more time load
> testing than you do just tossing up another server and using something
like
> kamailio or opensips to load balance...
>
>
> --
>
> Yossi Neiman
>
>
>
>
>
> On 11/19/2013 06:48 PM, Antonio Teixeira wrote:
>
> Hello Andre.
>
> Solving performance problems is always hard , but a few points to keep in
> mind :
>
> To my knowledge FS is pretty much a linux software , it works on windows
> by that i doesn't means it performs on windows ( please correct me if i'm
> wrong)
> The native DB for FS is Postgresql ( this should make you oracle admin
> more happy :) ).
>
> Try this :
> Linux Server Realtime Kernel + Last FS Stable + Postgresql ( tune it
well).
> Take care with ulimit on linux , its needed for the extra punch of
> performance
> Check if your "performance test" is real if you are pointing to 1500CPS
> for a two minute call you will have 180k calls before the first call gets
> disconnected,  if you are also testing RTP that means 180k * 2 ports
needed
> , The Kernel will start to complaining and you will run out of ports you
> will also run out of FD probably before you run out of ports on a not
tuned
> linux install.
> If the DB is on the same server it will also  stall since the FD will not
> be present for new connections to the back end.
>
> In terms of BW you will exceed 230000 kb/s add a few more kb/s for the DB
> ( Not a problem here i think)
>
> Another thing , no sip server ( please include all the nice names here
> like B2BUA , etc ) should ever queue anything.
> I have never read the code that handles this on FS but the way i would do
> it should be something like : can we sustain the load ? No in that case
> send 503 else proceed.
> This is done in freeswitch by this var
> https://wiki.freeswitch.org/wiki/XML_Switch_Configuration#max-sessions
>
> A more general approach:
> In SIp Servers you will generally ( if not running cpu intensive tasks,
> trancoding , tone detection , etc ), consume all the ram , Network IO
,than
> the cpu.
> For the guys that plan something big you should take a look at this :
> http://www.solarflare.com/High-Frequency-Trading
> Not by The NIC itself but because of this :
> http://www.openonload.org
>
> Without more data i don't think we can help , above all performance
> testing is a dark art you just need to keep on trying different things and
> above all have some type of monitor tool to understand what exactly is
> holding you up , maybe its no FS...
>
> Also 1500CPS is something seen in  carrier operations generally with a
> named SBC in front , keep in mind that FS is a great software but it
> doesn't hold against SBC's that use special processors like ASIC to
perform
> transcoding , sip parsing , tone detection ,etc etc
>
>
> Disclaimer : post written deep into the night...
>
> On 20/11/13 00:09, Andre wrote:
>
> Hi, I am testing out my FreeSWITCH  1.2.14 64 bit windows 8.1 on my
> Alienware Laptop with a very fast SSD and I7 processor with 8 GB DDR3 Ram.
>
>
>
> I have been using a pjsip test tool to test FreeSWITCH to see how it
> performs. Tell me if I'm wrong here, I believe if I sent "too many" calls
> to FreeSWITCH  that it queues up the calls and sooner or later uses all
the
> memory on my PC then just stops working.  I do have the CPS sent to 1500
> for testing but I never hit that high, I hit about 200 CPS.
>
>
>
> The CDRs that save to my database show it takes about 8 seconds from start
> time to end time. It does start out only taking milliseconds then it
> progresses to the 8 second mark.
>
>
>
> The calls are just going to FreeSWITCH then I'm sending back a 503.
>
>
>
> I have every mod turned off, logging set to 0, Core db going to a SQL
> Server and I even set FreeSWITCH to real time.
>
>
>
> Can anyone give me advise on what I can do to get better performance?
> I've also tested it out on my real server but in about 3 minutes of 50 cps
> FreeSWITCH uses up 3gb of memory and stops responding.
>
>
>
> I'm at a loss here on what else to do.
>
> Andre
>
>
>
> _________________________________________________________________________
>
> Professional FreeSWITCH Consulting Services:
>
> consulting at freeswitch.org <mailto: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
<mailto: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
>
>
>
> _________________________________________________________________________
>
> Professional FreeSWITCH Consulting Services:
>
> consulting at freeswitch.org <mailto: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
<mailto:FreeSWITCH-users at lists.freeswitch.org> 
>
> http://lists.freeswitch.org/mailman/listinfo/freeswitch-users

[The entire original message is not included.]


_________________________________________________________________________
Professional FreeSWITCH Consulting Services:
consulting at freeswitch.org <mailto: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
<mailto: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

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.freeswitch.org/pipermail/freeswitch-users/attachments/20131122/e530e8b0/attachment-0001.html 


Join us at ClueCon 2013 Aug 6-8, 2013
More information about the FreeSWITCH-users mailing list