Re: Freeway or Feeway?

Rohit Khare (
Wed, 22 Jan 1997 21:22:02 -0500

> Feeway or Freeway? Economics of the Internet*--Panelists: David Clark,
> senior research scientist, EECS; Hal Varian, economist, University of CA
> at Berkeley. Moderator: Robert Metcalfe, inventor of Ethernet, founder
> of 3Com Corp, vice president for technology at International Data Group.
> Technology and Culture Forum, 4pm, Rm 34-101.

You can find out more about the TaC Forum at

This talk promises to be quite exciting. I've actually talked to Drs.
Varian and Metcalfe separately over the last week about Internet payments.
Dr. Clark, as most of you know, is one of MIT's local protocol experts,
someone I'd like to do PhD research with if the opportunity arises.

Rev. Jane Gould introduces this talk and flags two forthcoming talks:

Internet Rights vs Internet Security with Barlow and Schiller
The Implications of the Internet Marketplace with Dertouzous

All 3 speakers are MIT grads; Metcalfe is incoming Alum Assoc President.
Bob's intro merely mentions that today's moderator is in no way moderate

"the moral and ethical implications of technology, today the Internet"

B: Problems: 1) Dave and Hal agree enough that they don't need moderation
2) it's not one of my strengths.

Dave has been at MIT since 1973, chair of the CSTB, which incidentally is
sponsoring the Rights Management workshop in Irvine I'm going to on 2/5-6.

Hal is an "actual" economist.

All three received their PhDs in 1973 ** intriguing...

Relatively uneditorialized minutes follow

D: "The Internet Intelligentsia", of which Dave is a founding member, and
from which I have since been thrown out ever since I started calling them
that... They are in a snit because these clueless newbies are trying to
define the future of Internet economics, politics, and culture. Which is
odd, since the II has shown no interest over the last 25 years in
economics, politics, and culture... Now that socialism has been discredited
in the real world, can we expect it to work in the virtual world?

Hal speaks: first, I spend so much time on-line, I'm now a virtual
[Interlude: all should know and love the infoecon mailing list]

Yesterday: single backbone, subsidized by NSF (~$12m/yr). 45% ftp, 15%
email, 10% logins, mostly ascii traffic.

Today: as of 4/30/95, goodbye NSFnet. many, many new backbones, NAPs, etc.
New applications: Web, video, java. Probably even *higher* rate of growth
now. --> self-supporting in principle (i.e. no one admits to making
money!). They're losing money on each packet, but intend to make up for it
in volume.

Overview of talk is to introduce problems facing key players: ISPs, Local
Exchange Carriers, broadband to home, backbone providers, interconnection,
QoS, multicast, accounting and billing.

ISPs range from AOL to neighborhood shops. cost structures: Equip 10%,
lines 30%, Operations 30%, sales (30%). Humans are 50%, the rest goes to
telcos. Why don't the telcos do it themselves? labor, regulated costs --
PacBell sez $750 per ISDN drop under *ideal* conditions. So retail
bandwidth is expensive to maintain; we'll see wholesale is cheaper without
user support problems.

Broadband to Home: ADSL, xDSL: Digital Subscriber Lines do digital over
legacy wiring (up to 1.5 MBps), but can only broadcast downstream. Quality
is horrible -- need all kinds of amps, filters, etc. It's shared bandwidth,
and usually asymmetric, can be come congested. ~$1000 per home. crosstalk
problems. Only 60-70% of homes are even physically close enough to COs to
make this work.

Aside: symmetric vs asymmetric b/w. Used to be a debate between "pushers"
and "home publishers". Not a big problem if home users can cheaply host
their content at the center, rather than serving from home. Only remaining
argument for symmetric is real-time, as in videoconf.

Local Loop Congestion. Voice call: 3.8 min avg; net call is 20.8 min. 10%
are over 6 hrs. Peak is 10PM for data, 11AM for voice. Abuses flat rate
plans, according to local exchange carriers. Bell Atlantic claims if 15% of
customers used the internet for 1hr per day, they'd have to double their
switches. So come on, why invest in POTS technology? make it cheaper to do

Cute demo: Microsoft had a demo Winsock that can aggregate modems: they
showed 4 POTS lines in 4 modems doing ISDN quality speed for $48/month
equivalent for $100s from PacBell ISDN. Heck, why not buy 8 lines and 8
modems? :-) Real problem is all-you-can-eat for $12 is not viable and
distorting decision pathways.

Backbone Providers. Rule 1: lighting up dark fiber you own is ~10x cheaper
than leasing it. Now, you can have SO much capacity, you can guarantee QoS
for some large customers; rest sit behind ISPs (and ISP's tech support
lines!). Aside: today 800 numbers are 1/2 of all long distance calls. Could
end up with a very concentrated markets.

[RK aside: what does it say that there's kind of a balance-of-bit-trade
that 1/2 calls cost the caller and 1/2 cost the callee? It's like the Kudos
model that depends on 1/2 bit consumption and 1/2 bit production to form a
closed-loop currency. See FoRK: ]

Interconnection. The business model here is really, really screwed up.
Gotta 1) share fixed costs, e.g. routing tables. Tempting to pass off
packets to a neighbor, like in the old NSF days where the NSF routed
everything 2) Usage sensitive costs e.g. QoS. 3) poor accountability, since
the end-consumer can't identify the culprits or incent them. 4) Settlement.
with what technology? no settlement systems exist today to compensate net
operators. Conclusion: Interconnection incentives are weird. Hal says he
has an example about how 2 parties can work ok together, but a third
destabilizes it -- downstream resale. Charter members of Commercial
Internet Exchange paid CIX, say, $10k/yr for peering services. Some ended
up reselling their "share".

Quality of Service. New applications demand more precise guarantees. How
can we handle this? separate networks (hmmm... isn't the data/voice split
where we came in??) RSVP -- yeah, but what does it cost to reserve it?
priority bits -- why should anyone turn on the low-priority bit?. Need
well-aligned incentives. How can you fold price-sensitivity into the
network, into the APIs.

Multicast. Actually, Hal is thinking mainly of multicast caching updates.
How can we pay for caching, provide proper incentives, etc. AOL is one
model: massive single cache. @home intends to deploy many, many
hierarchical caches to provide premium national service. Who pays for
caches? Who pays to select what goes in the cache. Both users and
publishers benefit from caching; how to split costs/benefits.

Hal hands it over to Dave at this point. "Internet: data flow or money
flow, or, Routing algorithms for money in the Internet"

We did a great job routing data, but now we're finding that we just can't
route money to some places in the Internet.

Today, the internet is not free. Find the money and follow it -- it starts
in your pocket, at the edge. Right now it's "regional" payments. From ISPs
to wide area IAPs. Then the backbones connect at "revenue neutral points".
[RK: cool, what a neat picture, paint the topological bit flux and find
NAPs only at the zero points. Why? because we only have peering,
settlement-free NAPs: they can ONLY support zero bit flux].

What's wrong? 1) the payment model is not driven by cost OR value. 2) it
offers poor incentives to various parties (users, provider, society). We
can even go beyond 'recovering costs' to implement value judgments.

Today, user behavior is affected by flat fees. It encourages use, which
encourages innovations -- want to keep that! -- but it encourages waste
(spammage, naive protocols like HTTP/0.9). Flat fees were critical to the
incubation stage of the Net. Experimenting ONLY happened because it was
free. [RK: how can we sponsor munchkins for free, too? Who will pay for the
first airdrop of wireless routers over Calcutta?] An alternative is usage
based perspective: pay by the bit or by the second. Bad, bad, bad because
it inhibits use, and makes bills more unpredictable for aggregate users.
For example, always-on type applications will really fail. Phone bills are
more predictable because human phone behavior can be modeled -- people
can't filibuster. But computes can. Dave left a laptop in the hotel plugged
in, and it magically dialed out at 2AM for 4 hours... Is there another

What about Provider behavior: there's no agreement about QoS, they just
sell "connections". Why not just load more users onto the same bandwidth?
People need motivation: complaints or money. fundamental: big users are
losing first because there IS a socialism in the net: when there's
congestion, the biggest pipes are reduced to the level of the mice.

What is the service model today? Best effort, or "send and pray". Powerful:
works over anything (!) and lets anyone build any kind of application,
encouraging adaptive apps. Uncharacterized performance -- perhaps
uncharacterizable? But in reality, users have different expectations; big
users will pay more. Internet weenies have never deployed anything except
computers and routers: they scrounged *anybody's* wires or clouds or

Dave's Excellent Payment Model. Users and providers negotiate a usage
profile, describing what to expect. You pay for the profile, the tag bit on
your flows; you can statistically test if a profile was met or not. If no
congestion, everyone gets what they want, since you're using marginal
facilities. When congested, costs finally show up: how to choose what to

Society and Payments. Today's telephone systems include cross-subsidies. We
hand out food stamps, someday we'll hand out bandwidth stamps, if Universal
Service erodes. Biz subsidizes homes; long distance subsidizes local; urban
subsidizes rural -- societal judgments enforced by regulators. Today the
net has none of this. Is there a right to the Internet? How much 'net:
modems, ADSL, etc? Role of schools, library freenets? which applications --
email, web, video? How do we achieve these -- for Nets, OR for phone
companies -- regulation, monopoly, taxes?

Return to the flux discussion: How do we locate the revenue-neutral NAP
boundaries? Why are they always at boundaries between wide area ISPs? Well,
when two big companies lawyers sit down to negotiate who owes whom what
funds, the techies scream, screwit, noone owes anyone anything. Even
counting packets tells you nothing. Remember money flows from the edges
inward -- that's not how packets flow. Packet flow is NOT value flow.
Instead, the big guys just go, 'uuuh, okay, we look equal, let's do a nap
and say nothing". Big guy vs little ISP: little guy pays up.

Problem: leads to a Procrustean value system. Different users have very
different VALUE and different willingness to pay. A hot public-access site
and could have the same bit-flows, but the value to the provider
is radically different. Commercial one is getting money for content,
adverts, etc. Public service one does best effort just putting a server up.
Who pays? [RK -- yep, Kudos!]. Same big-vs-small rule applies to
intercontinental connects. All the websites are in the USA -- if value in a
website accrues to the reader, Europe pays, but if there's value to the
publisher, USA pays. Small countries always end up paying for the link to
the big ones; we get 'downstream' connectivity to small countries for free.

On the phone, 800 calls offer one bit of discrimination between value
models. Leads to concrete proposal #2: expected capacity profiles. How can
participants express their share of the value? "Payment zones" for expected
capacity say how far they expect to 'throw' their packets - Disney will
cover the country; public service site will say it only goes 30 miles;
end-user has to 'suck' it up to is level. Right now we can only approximate
this with private nets -- Metcalfe's extranets.

Paying for Content. So far, Dave has only talked about bit-service. Not
application/user-level content. Two models: phone companies don't care at
all about content; Hollywood ONLY cares about content. Well, it's moot
unless we can actually demonstrate customers buying content for money? If
we do, though, can we route money through the internet for this too? [RK:
**Basic question of Information Economics is the same: who derives value
from the informations & how can we efficiently and honestly share that
information? ***]

Conclusions. Access-based pricing is probably untenable (flat-fee). But
usage pricing is not the only alternative. We need to match money to value
because there are already places the Internet is stalled without money

Lesson: we could upgrade all households to broadband for $100bn. Easy to
say no. How can we prove that there's enough revenue to justify the
investment? Viacom's blackmail scenario is scary. Dave told medical guy's
anecdote about 'infrustration': he thought with bandwidth into the home and
home monitoring, he could save 1% of $1Tn medical costs -- $10bn. But the
phone guys said there's NO WAY for medicine to kick in 10bn towards the
cost. There's no way to inject money into the systems.


Hal: micropayments are cool because there are two accounting models: phone
company central measurement vs postal service distributed accounting and
micropayment supports the second. Distributing the accounting to the
periphery is the Internet way to do it. micropayment may have more future
in transport payment than for content payment. Second point: universal
service drives me up the wall -- it costs more for phones in rural Maine,
and more for parking in Back Bay. It's just cost of living.

Bob: Should the Internet be free?

Hal: No. It's ill-posed: who should pay?

Bob: no, what if it's a compelling government interest to hand out T1s

Dave: Are we gonna throw in the PC? Don't confuse importance with equity --
even food isn't free or even equally subsidized for all. If you can prove
you're poor, you might get some assistance.

Bob: the FCC is about to decree that ISPs should offer discounts to schools
and libraries based on need, from the universal service fund.

Dave: it's a valid societal investment to jumpstart the investment

Hal: we don't know the future of this fund anyway.

Question from the audience about the lawsuits to open local phones to

Hal: the big money is the 4 cent per minute access fee for long distance
companies that data traffic was exempted as an 'Enhanced Service Provider'.
Internet phone really highlights the situation. FCC punted -- they don't
really know what to do yet. Some net services are competitive, some are

Question about Europe.

Hal: well, they're privatizing; it's difficult to regulate when the gov't
and phone company are the same. Most importantly, Europe is metered service
all the way down. In some sense, that's why the Net grew in the US, because
flat fees encouraged this growth.

Dave: remember that the internet is a ping-pong ball in the regulatory

Dave said more about his accounting-less scheme: just put traffic meters
that say 'is this packet in the profile' and color the packets in or out,
and the let the 'rest of the net' shoot out packets preferentially. [RK:
kind of like speeding tickets]. As you aggregate up, the statistics become
more reliable and predictable and we can write contracts between the big
guys. Or just recursively use the same scheme for interconnects.

[RK: I see a spot of interesting work in teasing out the trust models
behind Dave's work: after all, money is trust. Who has to believe whose
meters? What is the audit trail, or appeals process? it probably gets
Kerberos-like, with cross-authenticated domains]

Dave describes Dartnet, a T1-level experimental testbed between 8
cooperating Darpa sites. They're also doing mostly simulator work.