Index Home About Blog
From: (Fred R. Goldstein)
Subject: Re: How Will Routers Compete Against Digital Switches?
Organization: Digital Equipment Corp., Littleton MA USA
Date: Thu, 14 Oct 1993 02:57:28 GMT

In article <>
(Garrett Wollman) writes:

> I'd like to correct some of the misconceptions that Fred has been
> spreading.  I hate to do this, since Fred probably knows more about
> the physical side than most of us, myself included, but when one's
> honor is at stake ... (half a smiley)

I do enjoy these not-quite-flame wars ... In practice it's these
controversies tend to bring out a lot of interesting details.  ;=)

I said that you can't really do telephony over IP, since packet voice
is like teaching a pig to sing.

I hold to this position.  Packet voice isn't impossible, but it simply
can't compete with circuit-mode telephony.  There are two different
classes of serious problem.  One is technical, the other economic.

The economic problem is that packet processing (IP) requires the
router to look at every single packet header and figure out what to
do.  This volume sensitivity (routers are rated in packet/sec) is fine
with bursty traffic, but isochronous streams just beat up on the
routers mercilessly.  The "solution" may be cheap fast microprocessors
(Alpha AXP, anyone ;-] ) used liberally in routers, but in general a
circuit switch has the advantage, since you just set up the link and
let the bits fly, untouched.  On a dollar/bps basis, circuit switches
are a tiny fraction the price of routers.

There is also a bandwidth cost: In order to prevent packet drops of
asynchronous traffic (and I presume voice-over-IP would be mixed with
asynchronous data traffic), the links would have to be utilized at
less than 100%; isochronous bandwidth doesn't have this problem.  (It
just blocks additional calls when full.)  Bigger buffers reduce but do
not eliminate this, but they add delay.  This is an ATM problem: ATM
switches designed for mostly isochronous traffic have short buffers,
but are very inefficient for data.

The technical problem is delay-and-echo.  Packetization adds delay.
Delay makes echo more noticeable.  Telephone lines/sets, being
designed for the two-wire analog network, have echo, so your own words
come back to you a round-trip later.  With circuit, there is virtually
no delay in the switch, so "speed of light" is the only issue, and
echo is manageable as long as you don't have satellites.  With packet
you inevitably have long delay, so you need echo cancellation (more
cost, distortion), and even then IP-size packets tend to have too long
a delay for comfortable conversation.  StrataCom, for instance,
handles voice tolerably well in 21-octet cell-switched payloads, but
even that requires echo treatment.

Now these two problems interact.  You can overcome some loss, as
Garrett suggests, by using either Forward Error Correction or
retransmission.  The former requires additional bandwidth overhead
plus longer packetization delay; the latter requires round-trip times'
worth of additional delay.  For data that's not a problem, but for
telephony it would be terrible.  BTW, "broadcast" (one-way) voice,
which is delay-insensitive, does not have these problems, so indeed it
can be carried as data.  But that's a rather limited use compared to
the massive amount of two-way telephone traffic now carried

So let's let pigs oink, birds sing, and frogs croak.  Routers are
wonderful for data.  Circuit switches are wonderful for voice telephony.

Fred R. Goldstein
Opinions are mine alone; sharing requires permission

Date: 16 Nov 1997 03:39:13 GMT
From: fgoldstein@bbn.|nospam.|com (Fred R. Goldstein)
Newsgroups: comp.dcom.telecom
Subject: Re: The Internet Will Swallow the Phone System

In article <>, wollman@khavrinen.lcs. says...

>> The author did not explain how this "capability" will improve.

> Pity.  That's OK, I can.

But the argument is refutable.  Indeed, the whole "Internet will
swallow" thread seems to me to be amazing hubris amonst those on the
Internet side who simply don't understand the complexities of the
phone network.  Carrying CB-radio-quality voice, or even sunny-weather
hi-fi audio across the Internet is (literally) childs' play.  It's
only a teensy fraction of what the telcos have to worry about, and
usually do.

I started off on the "voice" side in the '70s, and was Telecom
Mgr. here at BBN in the late '70s, doing PBX stuff when the ARPAnet
NCC (NOC) was ours, and with my desktop VT-52 connected to host
computers via a Pluribus TIP.  Now I'm quite conversant on Internet,
packet and ATM matters, and speak both languages.  I do however see a
failure to communicate.

> Point 2: Statistical multiplexing, which arises out of the observation
> that not everybody wants to talk at once, applies to data networks ...

Of course.  It's natural there.

> Point 3: There are essentially three types of data services: elastic,
> adaptive, and isochronous.  Elastic services are insensitive to
> short-term delays in the network ...
> Isochronous services are at the exact opposite end of the spectrum:
> they depend on traffic reliably getting through the network at a
> constant rate, with packets arriving neither too soon nor too late.
> For a very long time, telephony-type applications were thought to fall
> into this category, but it's now understood by networking researchers
> that they are, in actuality, adaptive services, the third major type.
> An adaptive service is one which can operate under widely differing
> network conditions, and provides some amount of buffering at the
> receiving end which can be adjusted to provide as good a service as
> the delays in the network allow.  The prototypical ``Internet
> Telephony'' application, Van Jacobson's research vehicle `vat', was
> the first major application to develop this service model.

Ah yes, the Nineteenth Coming of Etherphone!  Indeed the idea that
voice is elastic is not new.  Undersea cables in the '70s used it to
conserve costly (analog!) channels.  Voice was stored in little
bursts.  Silent bursts were discarded and good ones were shoved onto
available channels, preceded by tone bursts (to identify the original
channel).  I went to work at DEC in 1980 just as they were installing
a TASI system, STC's COM-II, which could compress 31 channels onto 16,
all analog tie lines.  I think its packet size was 39 milliseconds.
It was supposed to save money, but in practice it was a horror show.
Then in 1986, StrataCom brought out their digital TASI, the IPX, which
used somewhat shorter packets and was somewhat less horrible, but not
without pain.

All in the meantime, the REAL cost of bandwidth was plummeting.  In
1980, a domestic US toll call probably cost Mother (AT&T Long Lines)
about 15c/minute (call it my educated guess), exclusive of the share
paid via Separations to the BOCs.  Then the glass began to go in.
Nowadays, after some inflation to devalue the cent, the call probably
costs more like two cents.  The Dime Lady costs more?  She's still
paying a nickel or more per minute to the LECs at each end of the
call.  Big users connect directly to their IXC and don't pay the
"originating" half, and "virtual private network" calls with no Bell
switches involved are routinely sold at well under a nickel a minute,
depending on volume.

It's pretty hard to shave much off of two cents.

So why does Internet telephony look so good?  Because nobody sees the
two cents.  They see the retail price, which includes billing,
marketing (you think the Dime Lady works cheap?), and huge "access"
charge payments to the local telcos.  It's cutthroat and not very
profitable, but very little of the cost is for bandwidth.  Internet
Telephony bypasses the billing system, so it looks cheap.

Legally (in the USA), IF you carry voice across state lines and feed
it INTO the local exchange (NOT into an ISP as a packet stream but as
a pure voice call into the LEC), then you ARE a long distance carrier.
Sticking a Cisco router in the middle and running newfangled forms of
TASI doesn't change things.  If you only use the local exchange to
call up an ISP as a data call, then of course the per-minute access
charges do NOT apply, and that leaves Internet telephony a big niche
market for recreational chatting amongst computer hackers.  But if it
rings a real phone line, Long Distance is Long Distance.

And frankly over the past 20 years the quality of LD has improved
astonishingly.  A call from Boston to New York in 1977 was hissy at
best, that "Long Distance" sound, and transcontinental calls were
half-duplex too due to echo suppressors.  Today a call between the USA
and Australia sounds almost local.  Internet telephony, with its long
packetization delays and low-bit-rate voice, with its dropouts and
"adaptive" (that's a euphemism) quality, harkens back to the bad old
days.  Sure, an LD company could offer it, but why, when fiber optic
64000 bps channels are cheap enough?

> Point 4: These three types of services interact in an interesting
> way.  Specifically, services which inject their data into the network
> at a constant rate, or effectively so, will always win a battle for
> bandwidth against elastic services like TCP.  This has two salutary
> benefits: first, and most immediately, people will actually be able to
> use the service, at least so long as the network isn't totally
> saturated.

That's salutary in the Swiftian "modest proposal" way.  Translated:
Voice is anti-social and drives data off of the Internet.  TCP follows
the Van Jacobson Slow-Start and backoff rules (if it conforms to
spec).  Voice-on-net doesn't.  That's why VON is worse for the
Internet itself than it is for the telephone industry!  I'm concerned
that too much VON will degrade the Internet's data performance,
causing too much congestion. The phone companies will fend for
themselves.  Or at least the smart ones will -- telcos who cry that
the Internet are "ruining" the phone network are missing the boat too.
(ooooh, the temptation to say "Bell Titanic" here is too great, but I
will try to resist that mixed metaphor ...)

> Second, in the long run, users will eventually notice that
> their network performance is getting sluggish, at which point some
> fraction of them will purchase a higher level of service, again
> providing an additional economic incentive to expand the capacity of
> the network.

By which point, data is clobbered, and doesn't have the option of just
picking up a normal phone!

Voice-on-net is a cute hack.  It's potentially useful for "intranets"
where there is private bandwidth, and for some discount long-distance
services (especially overseas), where its low bit rates might be
economical compared to the older gear telcos use.  But a
well-engineered circuit-switched telephone network is a thing of
beauty, not much appreciated by many Internet wonks but beloved of
millions of subscribers.  That market's not about to disappear.

Fred R. Goldstein   k1io   fgoldstein"at"   +1 617 873 3850
Opinions are mine alone; sharing requires permission.

Date: Mon, 01 Apr 1996 03:20:24 -0800
From: John Higdon <>
Newsgroups: comp.dcom.telecom
Subject: Re: Formal FCC Complaint Filed Against I-Phone

At 1:32 PM on 3/29/96, Patrick A. Townson wrote:

> This is a special bulletin received Friday morning regards the
> squabble between the telephone companies and the I-Phone people;
> the ones who use the software which allows voice communication
> via the Internet. The war has started! It appears the carriers are
> serious about getting rid of this Internet feature. A formal
> complaint has been filed with the Federal Communications Commission.

And I must say, this is the most pathetic filing to come across my
desk in many years. Out of the box, let me say that I feel that
Internet voice communication is a substantial misuse of packet
technology. And I agree that it is an inefficient use of network

But when a trade organization is moved to eliminate -- not competition
 -- but an alternative to its members' product, it is apparent that
many have lost track of what the free market is all about.

Some of the more ridiculous parts of the filing:

>     ACTA submits that the providers of this software are tele-
> communications carriers and, as such, should be subject to FCC
> regulation like all telecommunications carriers.

Followed by:

>     ACTA submits that it is not in the public interest to permit long
> distance service to be given away, depriving those who must maintain
> the telecommunications infrastructure of the revenue to do so,

I was unaware that providing an alternative to "long distance"
established an entity as a carrier. But beyond that, no one is giving
anything away.  The software company is charging for its software and
the Internet connections are presumably being paid for and the
equipment is presumed to have been duly purchased or leased.

The ACTA is upset because no one is charging anyone PER MINUTE rates,
and that strikes at the universal heart of the IXC gravy train. IXCs
have been nervous about the Internet for decades. Imagine, data -- in
any amount -- flowing at a fixed monthly cost. Up to now, it was just
"data". But someone crossed the line and took a leak in the IXC's
plush parlor.

What ever happened to the concept of finding better ways to do things?
The ACTA scarcely mentions any notion that voice communications over a
packet network is a kludge. That is not what upsets the organization.
This is:

> nor is it in the public interest for these select telecommunications
> carriers to operate outside the regulatory requirements applicable to
> all other carriers.

WHY is it not in the public interest? The Internet is no longer
subsidized by the government, so that argument is out the window. If
people are paying for a technology, who has the right to say that it
cannot be used in any lawful way possible?

>     ACTA's carrier members must be certificated and tariffed before
> the FCC and most state regulatory commissions in order to render their
> telecommunications service to the public.

So what? Why is this relevant? Internet voice communications do not
(necessarily) even use the services of ANY of ACTA's carrier members.
Who cares what that organization's members have to do?

> In addition, ACTA carrier
> members are subject to the requirements of the Communications Act of
> 1934, as amended (the "Act"), and various state laws and regulations
> which prohibit engaging in unreasonable practices and/or unduly
> discriminatory conduct.

And for good reason. Even with these regulations in place, IXCs are
screwing the public at every perceived opportunity. It is not
completely apparently why this statement was made, since it casts an
appropriately unfavorable light on ACTA's membership.

>     Entities, like those which are described hereinafter, which do
> not comply with or operate subject to the same statutory and
> regulatory requirements as ACTA's carrier members, distort the
> economic and public interest environment in which ACTA carrier members
> and nonmembers must operate.

Software companies are not subject to carrier statutory and regulatory
requirements. Why? They are not carriers. How more obvious can it get?
"Distort the economic and public interest environment" translated to
English means "customers might just find alternatives to our
over-priced services."

Here is one of my favorite excerpts:

> Absent action by the Commission, the new technology could be used to
> circumvent restrictions traditionally found in tariffs con cerning
> unlawful uses, such as gambling, obscenity, prostitution, drug
> traffic, and other illegal acts.

I see. Drug dealers are so illiterate that unable to use E-mail, they
would instead use Internet voice communications, presumably depriving
ACTA's members of even THAT revenue. Wheneven you run out of
intelligent arguments, toss in "gambling, obscenity, prostitution,
drug traffic" and all the other boogeymen.

But the crux of the matter is that what they propose is impossible:

>     Permitting long distance service to be given away is not in the
> public interest.  Therefore, ACTA urges the Federal Communications
> Commission ("the Commission") to exercise its jurisdiction in this
> matter and: issue a declaratory ruling establishing its authority over
> interstate and international telecommuni- cations services using the
> Internet; grant special relief to maintain the status quo by
> immediately stop the sale of this software; and institute rulemaking
> proceedings defining permissible communications over the Internet.

Nice try, but there is no "Internet". There is no main office; there
are no branch offices. It is impossible to "monitor" the Internet. It
would be impossible to define "permissible communications over the
Internet" simply because 1) new types of services are coming forth
daily; and 2) such regulations would be unenforcable. Since data can
be scrambled and encrypted in an infinite number of ways, even packet
sniffing on randomly selected routes would yield an empty hand.

What do we have here? We have a software company distributing software
that provides for utilization of a common technology: conversion of
voice to data; data to voice. Nothing illegal about that. And then we
have users installing that software on systems that have links to the
Internet. There is nothing immoral or illegal about that.

But ACTA thinks "there oughtta be a law ...". Let's see ... "Federal
Law prohibits the installation of this software on any computer that
has a connection to the Internet. Violation of the law may cause
damage to ACTA-members' Cash Cow."

I predict that our currect method for handling long distance (time and
distance charging) will eventually go by the wayside. I also do not
believe that "Internet Phone" is the vehicle that will cause that to
come about.  But it has served one purpose: we now know that ACTA is
not interested in ANY technological advances that threaten its status
quo. That is good information to have in the future.

John Higdon  |    P.O. Box 7648   |   +1 408 264 4115     |       FAX: | San Jose, CA 95150 |   +1 500 FOR-A-MOO    | +1 408 264 4407
             |             |

[TELECOM Digest Editor's Note: John, that was a great rebuttal. As
Jeff Buckingham has pointed out in the message just before yours, it
is not at all certain that *all* or even the majority of ACTA members
support the position the organization has taken. The meeting in
Phoenix should clarify exactly where the members stand on this, and
Jeff will be reporting it when he gets back. And I hope that when 
Jeff does attend the conference he will let it be known that this
action by ACTA has not gone unnoticed by the Internet community.  PAT

Index Home About Blog