[8772] in Commercialization & Privatization of the Internet
Re: One-way Internet?...
daemon@ATHENA.MIT.EDU (Larry Walker)
Sat Dec 4 11:17:13 1993
Date: Sat, 4 Dec 93 10:17:21 CST
To: com-priv@psi.com
From: walkerl@med.ge.com (Larry Walker)
Cc: barbiauxb@med.ge.com, ross@med.ge.com, hyperrick@aol.com
>I wrote:
>
>: So I'd be more excited by a $299 CSU/DSU than by a few more % points of
>: penetration...
> emv@mail.msen.com (Edward Vielmetti) replied:
>It's worse than that, actually. You can get a $299 high speed analog
>modem, <...> you'll either find
>that you async serial ports can't keep up with it (or) that you don't
>have a handy v.35 connector on the back of your computer.
>
>The other approach people have done is to deliver an ethernet interface
>instead of a serial interface, which then leaves you with either an
>expensive router (minimum $2000 these days per end) or an ethernet bridge
>that costs less (Combinet less than $1000) but is hard to scale up to
>100s of subscribers.
>
>Roughly no-one is making just the right sort of equipment - not
>that I have been able to tell at least - for this very obvious
>kind of application. The observation is that if you can get
>something that works for radiologists it will easily work for
>all of the corporate libraries, law offices, tool and die shops,
>engineering firms, software houses etc etc that have similar needs
>for internet access that's rapid, relatively bursty, and not
>grievously expensive.
>
>--Ed
Your last paragragh about broad applicability is exactly right: we need a
brouter-like device that plugs&plays in your average person's life, with a
hassle/confusion factor halfway between adding a CD player to your stereo
and hooking up a VCR with cable and premium channels, and which sells for
$299 at Radio Shack.
This then segues back into the recent brief discussion about Sprint's
Internet acccess service sans a home directory/mailbox/nntp-server and the
Imminent Death of the Net As We Know It (which I suspect could be true in
the short term; a lot of freely-provided net services _really_could_ get
swamped out of existence...)
The issue is one I've not seen discussed elsewhere. The concept is one that
my undergrad Urban Geography profs called "Break of Bulk": where do you
optimally locate, say a steel mill that needs iron ore from Lake Superior
and coal from West Virginia and limestone form <wherever> and also need to
deliver finished goods to their mix of industrial customers. You only want
to load/unload lake ore carriers ONCE! You get to use the Ohio river, but
barges should only load/unload once also. Then you have a mix of railroad
accesses and Interstate highway access for semis. Stir it all up in a
network model and optimize.
This kind of issue faces us as millions of people globally get an email
address with an @ in the middle. Where does Jane World Citizen go (plug
into, wireless over to) to get her email feed? Eudora, POP3? cc:Mail or
MS-Mail? GEIS BusinessTalk or MCI-Mail? Compuserve or Usenet for product
support? Where does she get her nntp news-server services?
How about home directories? When I'm in the office, I have a Sparc2 with a
home dir on the Unix servers at GE's National Support Center and read email
with Sun's OpenLook mailtool, and file articles to folders in my home dir.
When I work at home or on the road, I have a Mac with Eudora and get mail
by dialing in to a POP3 server behind a Livingston Portmaster, and file
articles on my Mac disk. A bit of quick-and-dirty shell scripting around
ftp lets me keep things resonably in sync, but we should wish this on 50
million of our fellow citizens?
Now where ought my home directory be? Where should I get my nntp-serve?
Note I said "serve" not "feed": Break-of-Bulk analysis of nntp is
instructive. Consider the reductio-absurdum approach: With increase
bandwidth everywhere, everyone on the planet can take a full Usenet
newsfeed right to their PC, 75 megs every night, so they can read the 3
postings in rec.elderberry.wine? Not.
So you keep the bulk up: Right now I have to go to the WELL go read rec.*
newsgroups, since my employer only feeds "work related" newsgroups. But I
don't want to pay cross-country phone rates to get to California. So I want
a local phone call to a dialin PPP server, who routes me to...? What is the
"natural scale" for an nntp server? With current 14.4kbps v.32bis modems
and conventional dial-in services, I'd want an nntp server in every "local
calling area". But if I have PPP via my TV cable provider, then I can get
my nntp service from anywhere on the net. So we put a giant news-server in
the middle of the desert somewhere, with mega fiber feeds and satellite
uplinks and everybody gets their nntp service from them. But that's the
Compuserve model, which is really mainframes redux.
The key, to me, seems to be much the same with respect to backbones or to
nntp services: We need to look at the issues in Internetworking which
correspond to the classic Breaks in Break-of-Bulk analysis: the intermodal
points where loading/offloading is incurred, the Natural Advantage points
provided by rivers, Great Lakes and harbors.
For nntp services and for backbone clouds, some key starting points might
be: Air-time vs packet-charges, blocking switches, and bandwidth
limitations between continents. I don't want to pay US$6-8/hr to read and
think and to compose this longish email message. But if I'm dialed in
through my local phone exchange, I AM consuming a scarce resource: a
connection through a blocking phone switch. Now if I can get my Internet
connection via TV cable, I pay either for monthly use of a given bandwidth,
or pay per packet consumed/produced, but NOT for connect time while reading
and thinking.
So there seems to be 2 viable models of email access: when at a "home
base", we'd want to utillze a "browsing" mode: transfer the
sender/subject/time window of an entire mailbox and let user select which,
if any, messages to transfer. When "intermittently connected" (i.e. through
any connect-time service with non-trivial charges), we batch-transfer
email, then drop the connection and let the user read and respond at
leisure off-line.
So this line of reasoning suggests for nntp service: Always take your news
serve from the same continent you are currently occupying! This may sound
trivial, but it's not. I (and other GE travellers) find that the easiest
way to get our email when travelling abroad is to dial back halfway around
the world by phone and log into either PPP/POP3 or to a cc:Mail PO in
Milwaukee. It seems fairly absurd to struggle through a transcontinental
modem linkup, and then have a dialin router make the packet connection the
last 8' over to the mail server! But it works, it's reasonably plug&play.
Ideally, if I'm at a conference in Amsterdam or San Francisco or wherever,
I should be able to dial a local exchange number and get an "IP-flavored"
connection (PPP, SLIP, whatever) that will allow me to access my POP3
mailbox "back home". This should work for those who maintain their home
mail service on a public provider. It can present a problem for those whose
corporate WAN is firewalled from the Internet, and can't access their mail
server from the Internet directly.
I throw these examples out to illustrate some of the ways we need to start
thinking about sensible ways to scale this Net thing up by a factor of
20-30x over the next couple of years.
Now I'd like to see someone continue this train of thought back to Eric
Fair's recent discussion of backbones as an ATM mesh with routers around
the edges for the second-tier access providers. That seems a natural
break-of-bulk point to me. One Internet cloud per continent? How many
access-provider-routers can share an ATM cloud until the exponential rise
in point-to-point routing proliferation swamps the machinery?
and so on..
