[11125] in Commercialization & Privatization of the Internet
Re: Options (was Re: What is an "Internet reseller"?)
daemon@ATHENA.MIT.EDU (Karl Denninger)
Tue Mar 22 06:11:12 1994
From: karl@mcs.com (Karl Denninger)
To: lcbginge@antelope.wcc.edu
Date: Tue, 22 Mar 1994 00:06:18 -0600 (CST)
Cc: matthew@echo.com, fair@apple.com, karl@mcs.com, com-priv@psi.com
In-Reply-To: <Pine.3.05.9403211930.A10822-d100000@antelope.wcc.edu> from "Bruce Gingery" at Mar 21, 94 07:47:26 pm
> On Fri, 18 Mar 1994, Matthew Kaufman wrote:
>
> > If I send 1MB of email from a Unix account I buy from some provider,
> > or 1MB of email by forwarding stuff via UUCP up to a service provider,
> > or 1MB of email by opening the SMTP connection myself, if its going
> > to some other CIX-connected host it uses EXACTLY THE SAME AMOUNT OF
> > RESOURCES.
>
> Actually, as I see it, that's not REALLY the case.
>
> 1. If you have direct IP, you may or may not move more data
> but what you DO use will use less resources of the
> various providers (or the same in some cases).
Not necessarily true. Direct IP users tend to do things like run Mosaic,
which NAILS their connection. A few of those is equivalent to a few FTPs,
(which is also going on)
> 2. If you use a UUCP connection, your provider can probably be
> a bit freer with scheduling OUTGOING data, but incoming
> targeted at you, whether ftp via net mailserver, list
> mail (or the list echo of a newsgroup, or even binnews
> packets), will not only have to be stored until you
> poll (unless bi-directional polling is furnished, which
> is NOT common).
Disk is unbelieveably cheap compared to real-time network bandwidth, and it
isn't "consumed" (ie: if you buy a GB disk you still HAVE a GB disk at the
end of the month. That T1 is a recurring cost).
> 3. If you have a "login" as opposed to UUCP or direct IP (SLIP/PPP)
> again, you MIGHT move less traffic, or you may utilize
> the MOST resources. Your value is likely decreased
> over BOTH of the above, but the resources used by the
> provider are likely the highest of the three compared options.
My experience is that a "login" and UUCP use MUCH -- MUCH -- less traffic.
In fact, the only saving grace is that once you have a packet mode
connection up, you tend to not need technical support (on the other hand,
the initial curve is a bitch for most folks).
> All of this presumes full network connectivity limited only by the
> type of service provided. Perhaps the best of all the worlds is..
>
> 1. uucp (via TCP) for incoming mail
> 2. pass-through straight IP for everything else.
>
> FOR THE PROVIDER
No. The best of all worlds in terms of users per connection is the shell
account. I can cram 20-30 people on a machine with a 56kbps backlink
depending on what they're doing.
I can't even get TWO 28.8kbps packet users on that same 56kbps line!
The reality is that anyone selling SLIP or PPP is automatically going to
need a T1 backfeed, regardless of how many people they have as customers.
The bar is MUCH higher in these cases.
> Though this eliminates some of the enhancements available in various
> UUCP implementations such as Z-modem based transfers, efficient bi-directional
> serial protocols, and the like. The IP connectivity eliminates much of
> the "user maintenance" requirements from the provider, at least as
> compared to "login" with home directory, much of the "auth" and "limits"
> setups, and the like. All of this is transferred to the consumer. Use of
> uucp TCP connects eliminates the need for multi-shell logins for most
> lines, and any term sets, shells (and special menued shells.) Much on-
> system storage. IF local newsgroup service is provided, it can be
> provided via simple INN and/or NNTP setups... no maintenance of rn and its
> successors. Storage requirements are limited to that required for any
> uucp store-and-forward system, but no maintenance of hand-keyed (hence
> betimes forgot) passwords, home directories, and NO need for support of
> novice users for on-system (rtfm) utilities.
Hahahahha..... :-) As someone who runs a business that does exactly this,
I have to say you're not even close to being right here.
> Connect times are apt to be longer than simple uucp, but shorter than
> logins, hence it hits a mid-point for serial resource use. With the
> problems some providers have had maintaining decent 'g' protocol
> transfers for simple uucp clients, the connect times may actually
> decrease, and the complaints certainly decrease.
Not true. I have statistics on our usage patterns to back this up.
> FOR THE CONSUMER
>
> Again, loss of Z- or bi-directional serial protocols for massive data
> movements will likely increase (depending on the alternative 'g'
> responsiveness) transfer times for E-mail, hence increase connect time
> somewhat. On the other hand, while connected, the link is not limited to
> JUST transfer of mail. Gopher, WWW, telnet to a handy Archie server (or
> use of a local Archie client), and pass-thru SMTP transfers can co-exist
> with the uucp connection. Advanced ftp clients (NEXTSTEP Yftp, ncftp,
> etc) can simplify and speed the access to remote files with a straight-
> through "download" as opposed to ftp to a login, then download with kermit
> or other facilities.
This is the problem. As soon as people get a net connection, the FIRST
thing they grab is Mosaic. Once that's running, get ready to have your
backlink line melted into slag.
> > "IP is Special" is a MYTH. Except for some arbitrary, unfounded beliefs of
> > some people who, now that they have spent their $10k, want to pretend
>> otherwise, there is NO REASON to differentiate between "user who occasionally
> > calls a Unix host", "user who occasionally sends out stuff via UUCP" and
> > "user who occasionally dials in and uses SLIP"
>
> My own two cents is that "IP is Special" is far from a myth. It's more
> bang for the buck (if provided at comparable connect rates) for BOTH the
> provider and the consumer. For the provider, it is more efficient use of
> resources. For the consumer it is potentially FAR more bang for the buck.
> The actual data volume is apt to be no more, or not MUCH more, and the
> local administrative problems are less.
Its more bang for the buck all right. But if you don't architect things
correctly it will murder your overall performance and lead to EXTREMELY
unhappy customers.
Multi-site, fixed IP number SLIP and PPP connectivity is NOT trivial to
make work correctly. Neither are the engineering and traffic monitoring
issues inherent in such a proposition. Nor is the billing and related
stuff. This is very much a "roll your own" area of the business right now.
Its a myth that the SLIP or PPP customer uses less resource. I can back
this up with facts. I'm not insane enough to be out here selling this
without putting in place the necessary tools to know what's going on with
our backbone!
> > They can all generate and receive bits which travel across the CIX router,
> > and as long as the connection is getting paid for, WHY SHOULD YOU CARE?
> > Once the CIX router and its support is paid for, the extra $10k per
> > "reseller" simply goes into Bill Washburn's pocket, as far as I can tell.
> > And simply MAKING UP reasons why more people ought to be required to
> > pay the $10k doesn't go any farther towards convincing me that its going
> > to a good cause.
> >
> > -matthew
>
> Bruce Gingery lcbginge@antelope.wcc.edu
--
--
Karl Denninger (karl@MCS.COM) | MCSNet - Full Internet Connectivity (shell,
Modem: [+1 312 248-0900] | PPP, SLIP and more) in Chicago and 'burbs.
Voice/FAX: [+1 312 248-8649] | Email "info@mcs.com". MCSNet is a CIX member.
