[Asterisk-Users] MTU and Voice Delay (latency??)

[Rusty Dekema]

On 1/9/06, Terry H. Gilsenan <thg at interoil.com> wrote:
>
>
> I have phones in the US conencted to and Asterisk box in .AU, the ping
> time averages 230ms, (ADSL at both ends) and the call quality is just
> fine.
>
> The problem will be if packets are being dropped, of if one or other of
> the end-points is getting saturated. I have Linux/IPTables with ToS and
> QoS giving voice data the highest priority.
>


That is very true when it comes to actual voice quality. But the original
poster's users are complaining about latency.

If one of the end-points is getting saturated, then that fact should be
reflected in the round trip ping times, as the icmp (ping) packets will have
to sit in the buffer of the saturated end's router/bridge until its turn
comes to get sent down the line. A Linux IPTables machine with ToS and/or
QoS at each endpoint (or even just the congested endpoint) can certainly
help solve this kind of a problem; in fact I use one myself. But the problem
itself is usually detectable by abnormally high ping times from endpoint to
endpoint and a sudden jump in traceroute ping times along the route once the
saturated interface is reached.

If the round trip time across the Internet is 230 ms as in your case, or
(100 ms) in the case of my earlier rule of thumb, one-way Internet time
would be 115 ms (50 ms) and wouldn't be considered a problem. But, an
internet one-way trip time of 115 ms (50 ms) is not normally going to
provide a mouth-to-ear voice path time of 115 ms (50 ms), since additional
delays will be caused by the jitter buffer if one is in use, as well as by
any echo cancellation that may be in use. In my experience, if there is much
jitter at all on the Internet path, voice quality will degrade unless a
jitter buffer is used. But, the worse the jitter is, the larger the jitter
buffer needs to be and the more latency will be introduced by it. So you've
essentially got a tradeoff between "voice quality" (i.e. lack of jitter +
lack of packet loss) and latency.

It does not take much jitter nor all that much latency to cause degradation
in the quality and mouth-to-ear delay of VoIP calls. You could easily have a
line that would never give you any problems in a web browser or even an ssh
session, but that would severely interfere with the quality of a VoIP call.
In this case, if you can reduce the Internet latency and/or jitter between
your two endpoints, you can improve voice quality of and/or reduce the delay
present in your VoIP calls.

-Rusty
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.digium.com/pipermail/asterisk-users/attachments/20060109/6e8503f7/attachment.htm