Cell Relay Archive

[James Carlson <carlson@ironbridgenetworks.com>]

albert.e.manfredi@boeing.com writes:
> > As the original poster noted, it certainly is an issue for the folks
> > paying for the links.  If you're not footing the bill, then I suppose
> > it is less important.  (TCP over ATM without IP?  References?)
> 
> Right, but the whole overhead issue is, in my opinion, not put in
> perspective.
> Let's say you are really planning on using IP for voice. Compare that with
> using ATM for voice, over AAL1. What are the overhead comparisons?

Ok, let's do that, then.  G.723.1 encodes 30ms (240 samples) of voice
into ten 16-bit (20 bytes) code words in 5.3Kbps "toll-quality" mode.
If this is transmitted over UDP/IP (28 bytes of header) as
packet-over-SONET (8 more bytes of overhead), the result is a lousy
36% efficiency (20/56) before considering silence suppression.

But so what?  Efficiency of low-speed applications like voice just
isn't an issue.  Efficiency shows up as an issue only when you're
looking at data in the aggregate.  Non-voice data transmission already
surpasses voice in public networks, and data is growing in volume at
anywhere from 100% to 900% per year while voice is essentially flat.
By the time CLECs deploy voice over IP, voice will be just a niche
data type.

(If you choose to compare AAL1 against voice over IP, things are much
better for IP.  AAL1 requires six cells at 318 bytes to transmit that
same voice data, making voice over IP five and a half times more
efficient.  If you try to compress to make the comparison fairer, then
AAL1 has at best a marginal efficiency improvement, since it is forced
to use 48 byte cells whether or not it needs to; which means that the
same 30ms sample that takes 56 bytes on an IP link takes 53 bytes on
an ATM link.)

> Since this sort of application would most likely use UDP/IP over Ethernet, the

Ethernet?  How did that crop up?  Did we start comparing SONET against
Gigabit Ethernet?  That would be a fun discussion, but I don't think
we were doing that.  We were looking at voice on point-to-point
links, not broadcast media.  PPP is used for packet-over-SONET, and
its overhead is nominally 8 bytes.

> If you use AAL1 for voice, over SONET STS-3c, for a similar (but more
> credible) QoS level, all you need at the 155 Mb/s trunk level is 14.6 percent
> overhead.
> 
> So again, you can't fool mother nature. If you'll mix interactive voice into
> your IP network, you will have to ensure that the data services don't get in
> the way of the voice services. And to do that, the data service efficiency
> will be reduced. In the actual voice part of the scheme, the small cell
> overhead of ATM shows its worth. In ATM, the signaling evolution made it
> possible to do without all the packet overhead IP requires.

"It depends."  If voice continues to shrink in relative market size,
then quality-of-service, at least for voice, is simply a red herring.
Flagging a few percent of your data as "highest priority" essentially
has no effect on best-effort data traffic, and causes zero loss of the
voice traffic.

And, once again, ATM insists that you pay this performance tax,
whether or not the "quality of service" features are important to you.

-- 
James Carlson, Consulting S/W Engineer  <carlson@ironbridgenetworks.com>
IronBridge Networks / 55 Hayden Avenue  71.246W    Vox:  +1 781 372 8132
Lexington MA  02421-7996 / USA          42.423N    Fax:  +1 781 372 8090
"PPP Design and Debugging" --- http://people.ne.mediaone.net/carlson/ppp