[11105] in Commercialization & Privatization of the Internet
Re: What is an "Internet reseller"?
daemon@ATHENA.MIT.EDU (Karl Denninger)
Tue Mar 22 02:35:26 1994
From: karl@mcs.com (Karl Denninger)
To: fidelman@civicnet.org (Miles R Fidelman)
Date: Tue, 22 Mar 1994 00:58:42 -0600 (CST)
Cc: com-priv@psi.com
In-Reply-To: <Pine.3.89.9403190816.A11515-0100000@world.std.com> from "Miles R Fidelman" at Mar 19, 94 08:50:46 am
>
> > >From: "Erik E. Fair" (Your Friendly Postmaster) <fair@apple.com>
> >
> > >Why do some of you persist in believing that handing out IP addresses
> > >makes it qualitatively different? Bandwidth is bandwidth.
> > >
> > >This whole mess comes about because NSP's will not engineer their
> > >networks to carry the total aggregate inputs - too expensive. So, they
> > >guess at an average usage per customer, and then frantically restrict
> > >anything that violates that assumption, with lots of handwaving.
> >
>
> If the issue really is bandwidth, its not immediately obvious to me that,
> given a T1 Internet pipe, a collection of SLIP users eats up more
> bandwidth than a large public access Unix host. So... can anybody supply
> some real operational statistics on average traffic characteristics for:
>
> i. some public access hosts
>
> ii. some SLIP resellers
>
> There's nothing like real data to liven up a religeous arguement :)
A SLIP customer loads the network infrastructure MUCH more heavily than
the host user.
Here's the reasons:
1) The SLIP customer reads news over NNTP. NNTP is effectively telnet
in a package. This means lots of small packets, which are horribly
inefficient and harder to route than big packets.
The host user reads news over either local spool (no net load) or
over NFS (big packets, and lower overall load). Anyone doing NNTP
in this situation should look at the issues here and consider
switching. In many cases it can be a huge win.
2) SLIP customers run Mosaic a lot. Mosaic is a line burner. Host users
can't run it at all.
3) SLIP customers tend to "ping" on things (like mail servers) to check
for new traffic. Host customers generally don't do that, although
they may load a machine heavier.
4) On average, a SLIP session lasts significantly longer than a host
session. This does ugly things to your line usage models; we had
to adjust ours when we brought SLIP and PPP accounts online here
to account for the different loading pattern. Note that despite
our flat-rate shell accounts, our loading pattern there has been
stable on averages for over a year!
My observations are that SLIP customers load the network infrastructure
heavily, while host customers load the hosts heavily. Network
infrastructure tends to cost more money than host machine power and disk
space. On the converse, a SLIP customer, once set up, generally doesn't
cause trouble tickets to be generated (but the curve on initial install is
nasty). Host customers are easier to set up, but tend to be a support
load all the time.
Its a trade-off. Much depends on what you have more of, where, if
anywhere, you have excess capacity, and what kind of choices you made when
you purchased equipment and, in the case of Internet backlinks, services.
--
--
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.
