Karl Auerbach comments on VOIP (yes he does have a life outside of ICANN. ;-) djf)

[Dave Farber <dave () farber net>]

------ Forwarded Message
From: Karl Auerbach <karl () cavebear com>
Reply-To: Karl Auerbach <karl () cavebear com>
Date: Sat, 15 Mar 2003 17:52:22 -0800 (PST)
To: Dave Farber <dave () farber net>
Subject: Re: [IP] some comments on Review of Vonage's VoIP broadband phone
service


Over the last few months I have been building an end-to-end test lab to
evaluate the behaviour of voice-over-IP systems and phones under
controlled degrees of network impairment.  (Yes, I do have a life outside
of ICANN. ;-)

See the early report:
    http://www.iwl.com/Products/maxwell/VoIPReport.html

and somewhat more product oriented:
    http://www.iwl.com/Products/maxwell/Maxwell_for_VoIP.html

What we've done is to build a box (we call it "Maxwell" after James Clarke
Maxwell's famous 2nd law of thermodynamics daemon).  Max allows us to do
stateful impairments of the traffic flows - losses, duplications, delays,
reordering, packet rewritings, etc.  Around this we've built various test
rigs to let us inject analog voices and tones into phones and measure the
analogue coming out of the other phones.

This is somewhat different than other kinds of testing, mainly used with
POTS phones, in which the phones themselves are not part of the test
system.

But with Voice-over-IP the quality of the RTP/RTCP implementation in the
phones is one of the most critical aspects of voice quality.  I spent
several years at Precept Software/Cisco working with RTP/RTCP on IP/TV, a
high quality audio/video application and I can attest first hand that even
though the RTP/RTCP RFC's are not very thick, they are quite hard to
implement well.

Anyway, we've got a couple of dozen phones - soft and hard, SCCP and other
call setup protocols, including the Cisco ATA 186 that Vonage uses.
Although various setup protocols are used, all these phones ultimately use
RTP/RTCP for the actual voice transport.

We're still pretty early in our work on this, but the results are, to put
it euphemistically, quite interesting.  And its hard to nail things down
into categories - there are a lot of adaptive algorithms that implementors
use to handle jitter (variation in transit delay) and sharp changes in the
perceived jitter (i.e. the variation in the variation of transit delay.)

We've also noticed some more subtle effects - all voice-over-ip phones
induce at least a slight delay (some home-brew softphones induce a couple
of seconds of delay).  At a certain point (the ITU says around
150milliseconds of delay) people start shifting to walkie-talkie mode of
speach rather than natural conversation.  We've noticed that before that
happens, there is a potential for the parties to start to believe that the
other person is responding slowly.  This may lead to false conclusions
that the slow person is under the influence.

When evaluating voice quality one has to be careful to compare apples with
apples and oranges with oranges - virtually every VOIP phone is software
driven and the phone's behaviour will vary from software release to
software release.  In addition, the specific configuration settings - such
as suppression of packets during periods of silence - can have an impact
on quality.

I'll be at the IETF in San Francisco this coming week if anybody wants to
get together and talk about this stuff.

        --karl--




------ End of Forwarded Message

-------------------------------------
You are subscribed as interesting-people () lists elistx com
To manage your subscription, go to
  http://v2.listbox.com/member/?listname=ip

Archives at: http://www.interesting-people.org/archives/interesting-people/