asterisk users - Feb 2007 - Diagnosing poor call quality

Greetings list,

We have an issue with call quality at 2 sites where the users (4 Elmeg
IP290s at one site, 2 SPA942s at the other) do not have an asterisk box
on-site. Each site has an 8mb down/448k up ADSL connection and the phones
connect via SIP to an asterisk box in a datacentre using g729.

The asterisk box in the datacentre connects to our other asterisk boxes
providing pstn connectivity via IAX2. Latency between these boxes is between
1 and 2ms. The ADSL connections to the client sites are all consistently
delivering latencies of sub-25ms to the datacentre and there is traffic
shaping on that connection to give priority to any traffic from the phones'
IPs.

Comments from the users at these sites are as follows:
"call sounded like a dalek and I couldn't make out anything the caller
was
saying"
"the phone on my desk is breaking up so badly it's virtually
unusable"
"calls sound like they're breaking up with metallic background
noises"

We have quite a few customers with asterisk boxes on-site (with phones
connected to them via the LAN) using ADSL connections from the same
supplier, and are not having these issues with them.

canreinvite=no and nat=yes are set on all these devices, since they are
behind NAT. Each device re-registers with asterisk every 5 minutes to
prevent any possible NAT state timeouts.

Any pointers/places to look for potential problems would be much
appreciated.

Regards,

Chris
-- 
C.M. Bagnall, Director, Minotaur I.T. Limited
This email is made from 100% recycled electrons

Chris Bagnall wrote:
> Greetings list,
> 
> We have an issue with call quality at 2 sites where the users (4 Elmeg
> IP290s at one site, 2 SPA942s at the other) do not have an asterisk box
> on-site. Each site has an 8mb down/448k up ADSL connection and the phones
> connect via SIP to an asterisk box in a datacentre using g729.
> 
> The asterisk box in the datacentre connects to our other asterisk boxes
> providing pstn connectivity via IAX2. Latency between these boxes is
between
> 1 and 2ms. The ADSL connections to the client sites are all consistently
> delivering latencies of sub-25ms to the datacentre and there is traffic
> shaping on that connection to give priority to any traffic from the
phones'
> IPs.
> 
> Comments from the users at these sites are as follows:
> "call sounded like a dalek and I couldn't make out anything the
caller was
> saying"
> "the phone on my desk is breaking up so badly it's virtually
unusable"
> "calls sound like they're breaking up with metallic background
noises"
> 
> We have quite a few customers with asterisk boxes on-site (with phones
> connected to them via the LAN) using ADSL connections from the same
> supplier, and are not having these issues with them.
> 
> canreinvite=no and nat=yes are set on all these devices, since they are
> behind NAT. Each device re-registers with asterisk every 5 minutes to
> prevent any possible NAT state timeouts.
> 
> Any pointers/places to look for potential problems would be much
> appreciated.
This should be a FAQ. Set the RTP packet size on the SPAs to .2 instead 
of .3

The advertised datarate (8mb/448k) are the speeds at which the circuit 
between the customer and the central office is clocked and has no 
relationship with *effective* throughput.  At the central office are 
*shared* facilities than connects each DSL connection with the network, 
and over subscription to these shared facilities cause congestion. 
Also, there is no QoS on the Internet, and congestion anywhere between 
the end points will cause poor call quality.

Disclaimer: The following information is several months old--I've since 
moved my customers away from Qwest DSL.

Here in Denver we have Qwest DSL "service" from a central office where
the effective throughput drops to dialup speeds during the day.  Regular 
web/email users don't usually notice packet loss because dropped packets 
are recovered by the TCP protocol.  For VoIP on UDP, however, the call 
quality suffers to the point of being unusable (clicking, popping, and 
dropouts).

Furthermore, Qwest doesn't have Denver peering with the "rest" of
the
Internet.  To leave the Qwest network, connections typically go to DAL, 
LAX, or SFO on congested circuits.

So beware of VoIP over DSL.  Your users need to be aware of the 
tradeoffs between the cost of DSL vs. T1 and the effect on call quality.

Chris, if your customers are in the western US then please contact me 
about dedicated circuits.

Chris Bagnall wrote:
> Greetings list,
> 
> We have an issue with call quality at 2 sites where the users (4 Elmeg
> IP290s at one site, 2 SPA942s at the other) do not have an asterisk box
> on-site. Each site has an 8mb down/448k up ADSL connection and the phones
> connect via SIP to an asterisk box in a datacentre using g729.
> 
> The asterisk box in the datacentre connects to our other asterisk boxes
> providing pstn connectivity via IAX2. Latency between these boxes is
between
> 1 and 2ms. The ADSL connections to the client sites are all consistently
> delivering latencies of sub-25ms to the datacentre and there is traffic
> shaping on that connection to give priority to any traffic from the
phones'
> IPs.
> 
> Comments from the users at these sites are as follows:
> "call sounded like a dalek and I couldn't make out anything the
caller was
> saying"
> "the phone on my desk is breaking up so badly it's virtually
unusable"
> "calls sound like they're breaking up with metallic background
noises"
> 
> We have quite a few customers with asterisk boxes on-site (with phones
> connected to them via the LAN) using ADSL connections from the same
> supplier, and are not having these issues with them.
> 
> canreinvite=no and nat=yes are set on all these devices, since they are
> behind NAT. Each device re-registers with asterisk every 5 minutes to
> prevent any possible NAT state timeouts.
> 
> Any pointers/places to look for potential problems would be much
> appreciated.
> 
> Regards,
> 
> Chris

Check from the sites in question using "testmyvoip.com" or whatever
the
site is called.

In the UK I found that some strange things sometimes happen.  At one 
point I was sure that BT were perhaps misclassifying IAX packets as 
P2P... However, not had a problem with SIP.

Beware that ADSL uses vastly more bandwidth than you expect on small 
packets, eg if you are classifying using a cheap router then you 
probably need to at least half your claimed bandwidth in order to make 
the prioritisation work correctly.  I added some (hack) patches to fix 
the linux calculation for HTB on the linux QOS list a year or two back.  
If you have a linux router you could use those to improve the 
calculation quality for QOS - or else I found a Draytek router does 
impressively well at getting it right for small sites...

Very likely you will find that the issue is variable jitter on the 
line.  The link above should help you figure this out

Good luck

Ed W