[asterisk-users] stucked calls in asterisk 1.4

[Alex Samad]

On Thu, May 28, 2009 at 10:49:38AM +0200, Stefan Schmidt wrote:
> 
> David Backeberg schrieb:
> > On Wed, May 27, 2009 at 1:49 PM, Stefan Schmidt <sst at sil.at> wrote:
> >> all server are in one rack in our datacenter and are connected to an HP
> >> Procurve 2650 switch, which has been setup around 3 months ago, cause of
> >> the old switch died silent in the night.
> >>
> >> all server had two interfaces and i have allready tried to route the
> >> traffic between the pbx and the routing server over the second
> >> interface, where database requests normally run. But this didnt solved
> >> the problem too.
> >>
> >> i will try to increase the UDP buffer size in the linux kernel, maybe
> >> this will take some affect.
> > 
> > I will say that asterisk-1.6 is supposed to have a better SIP stack
> > than 1.4. Perhaps the difference in performance will help you.
> > Specifically, check out:
> > 
> > http://svn.digium.com/svn/asterisk/branches/1.6.0/CHANGES
> > http://svn.digium.com/svn/asterisk/branches/1.6.1/CHANGES
> > 
> > I'd recommend you go to at least 1.6.1.* series
> 
> i will have a look at 1.6.1 version but the problem is a network/kernel
> problem.
> 
> i´ve tried to set the udp rcv buffer to 1mb but that didnt change the
> problem.

Hi

I am new to asterisk so my suggestions might be a bit silly.

Why not setup a iax2 connection bettween the asterisk servers, because
its a lower overhear and more efficient.

> 
> what i can see is that the asterisk itself sometimes lag so no call can
> be started, but running calls wont be broken.
> 
> in netstat -su i see a big amount of udp packet receive errors when it
> lags. Like 600 to 800 pakets per second.

if you are seeing packet loss, have you check the stats on the cards and
the ports on the switch any sort of loss might indicate a faulty
card/port

you could also try some network testing with iperf by default it tests
udp.

the other thought is keep going the way youwere add another nic card to
server a such that you have 

eth0 for your sip clients
eth1 for router server b
eth2 for router server c

it would just take a bit of routing magic.  The thing to watch then is
the bus these card are on if they share the same pci bus.  But you are
not really moving around much data.

Try linux-kernel mailing list or do a  google for udp buffer tunning
from memory there is the kernel option which is the abs max, but the
program must also select a buffer amount which is  usually the default
value which is not the max value.

another thing to look for (you haven't mentioned you nic type or kernel
version), but I had a realtek 8168b which would sort of work with the in
kernel driver (8111c) on any kernel < 2.6.28. After 2.6.28 the in kernel
driver worked really well. system would be under load packets would be
come corrupt

for example

ethtool -S eth0
NIC statistics:
     tx_packets: 75371387
     rx_packets: 970026891
     tx_errors: 0
     rx_errors: 0
     rx_missed: 0
     align_errors: 0
     tx_single_collisions: 0
     tx_multi_collisions: 0
     unicast: 969467810
     broadcast: 201337
     multicast: 357744
     tx_aborted: 0
     tx_underrun: 0



> 
> in syslog i cant see anything that would cause asterisk to react like this.
> 
> anyone an idea what can cause this problem?
> 
> best regards
> 
> steve
> 
> _______________________________________________
> -- Bandwidth and Colocation Provided by http://www.api-digital.com --
> 
> asterisk-users mailing list
> To UNSUBSCRIBE or update options visit:
>    http://lists.digium.com/mailman/listinfo/asterisk-users
> 

-- 
Documentation:
	Instructions translated from Swedish by Japanese for English
	speaking persons.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: Digital signature
Url : http://lists.digium.com/pipermail/asterisk-users/attachments/20090528/f1f5a5f9/attachment.pgp