Cell Relay Archive

[fgoldstein@bbn.com (Fred R. Goldstein)]

In article <1995Apr27.140227.20553@nb.rockwell.com> manfredi@engr04.comsys.rockwell.com (Albert E. Manfredi) writes:

>On the other hand, the ATM Forum is apparently going to address _real_
>use of ATM for voice, in which, I would expect, an SVC is assigned to
>each telephone call, at typically 64 Kb/s bandwidth.

>I'm puzzled about your thinking that this would be "an issue." After all,
>the way ATM does signaling, it seems tailor-made for just this sort of
>service. It is, after all, a telco invention, yes? I think the only real
>issues are that people want this sort of ATM service to use the same
>Q.931 signaling as ISDN, rather than invent something else. And ATM does
>not do Q.931 now.

It's a "big duck" issue.  ATM was created by telcos as an attempt to do 
everything and more.  A duck can fly, swim and walk, but is no eagle, fish or 
cat; ATM carries all sorts of traffic, but not always as well as 
more specialized alternatives.

ATM signaling (Q.2931) is derived directly from Q.931, so that's not the 
problem; in a sense, Q.2931 is really just an extension of Q.931 with some new 
ideas thrown in.  It was originally intended to be the short-term solution for 
ATM, but nobody's working very hard on a successor.  ATM's weakness for voice 
(SVC per call) is really one of competitive necessity:  STDM (N-ISDN) does it 
better, so why waste much effort on it?   ATM's fault here is that it 
introduces packetization delay, about 6 ms. for cells filled with PCM, and 
delay is the bane of telephony.  Plus you need a little build-out delay to 
de-jitter the incoming cells.  So it can't sound quite as good as STDM.
If you work at it, you'll end up with something almost indistinguishable in 
quality from STDM, but probably at higher cost.

That's why it seemingly makes sense to use T1 circuit emulation to share ATM 
pipes between PBX trunk groups and data.  It minimizes delay and keeps things 
simple, leaving call intelligence in the arena where it is well understood.

Now there are potential optimizations by doing SVC voice, so it's not 
perpetually hopeless; it's just not likely to set the world on fire very fast.
The sad thing, of course, is that it was just this application that the short 
cells were designed for; data folks would have preferred 64 or even 128 octet 
payloads, but SVC voice is affected too much by the packetization delay.
___
Fred R. Goldstein   k1io    fgoldstein@bbn.com
Bolt Beranek & Newman Inc., Cambridge MA  USA   +1 617 873 3850
Opinions are mine alone; sharing requires permission.