[9896] in Commercialization & Privatization of the Internet
Re: bill text draft 2: Telecommunications Competition Act (fwd)u
daemon@ATHENA.MIT.EDU (Karl Denninger)
Wed Jan 26 01:33:31 1994
From: karl@mcs.com (Karl Denninger)
To: ittai@ans.net (Ittai Hershman)
Date: Wed, 26 Jan 1994 00:14:19 -0600 (CST)
Cc: karl@mcs.com, com-priv@psi.com
In-Reply-To: <CMM.0.90.2.759542971.ittai@shemesh.ans.net> from "Ittai Hershman" at Jan 25, 94 07:09:31 pm
> I would <prefer> to use the CIX transit where possible, and the NSF transit
> only where <necessary>. The routing tables may say otherwise, but that's
> not my doing -- all I can do is announce the routes I have, and <ask> MERIT
> to announce the routes to the PRDB for the NSF. The weights that MERIT
> attaches to those routes has a large part in the "best cost" path that is
> taken by the packets, does it not?
>
> ANSNet policy routing supports primary, secondary and tertiary paths.
> Technically, your network could be announced as 1:1957 (CIX-AS) 2:1800
> 3:1240 (Sprint AS's). Because of the way Internet routing works
> (nothing to do with ANS, Merit or NSF, just the facts of IP routing)
> this would mean that as long as connectivity were available via the
> CIX, packets destined for your network would be routed via AS 1957.
Which is, if you want to be pedantic about it, ANS' and MERITs choice.
In a perfect world routes would be computed on best path with no political
considerations. However, there <are> political considerations, aren't
there? You have, on this list, said "the CIX model is broken". Ok, so
what do you propose instead? Settlements? I don't like <that> model, and
I <particularly> don't like the model coming from ANS, which has ended up
with all the EDU sites by defacto monopoly positioning as a result of
the "cooperative agreements" granted by the NSF and the US Government --
even though, today, the charge that this is "research into IP connectivity"
is absolutely bogus -- any of those .EDU sites could attach via Alternet,
Sprintlink, PSI, or even, yes, ANS. No "test bed" required; this is
commercially available, off-the-shelf technology. Hell, I can buy T3/HSSI
interfaces for my AGS+ right now and put up T3 lines routing IP traffic
tomorrow! (Well, ok, I would have to "fast-ship" those boards from CISCO
and they might take 5 days to get here. The point still stands; this be
off-the-shelf technology today.)
OF COURSE ANS wants settlements. They would result, today, in a net-inward
cash flow to ANS, a net drain to everyone else. That makes perfect sense.
If ANS had fought for its market share in the world of public competition,
I might even agree with you. However, you have not done so, and thus I see
that argument as nothing other than self-interest fueled by a bankrupt
information highway policy at the federal level.
5 years ago this was a technology which wasn't commercially available and
reliable. 3-4 years ago <it was>; I was buying connections from Alternet
then, and they certainly weren't "testing" anything. That was production
quality IP connectivity from a commercial provider. Period.
The rest of the world doesn't like the idea of FTPing somewhere and being
greeted with a "VISA:" prompt. Which is, by the way, exactly what I
believe a settlement policy will lead to.
I also note that your "T3 network" has a <net 75ms transit time> (as
reported by traceroute) right now from Washington DC to San Francisco,
or about what I would expect (and have measured when I have built national
networks) from a <56kbps line> going the same distance. T3? Well, perhaps
in throughput (I can't measure that without loading down people's links,
and I'm not about to do that) but certainly not from a latency standpoint.
> However, because the CIX router can only be used to reach other CIX
> member customer networks, this would mean you would be cutoff from a
> large part of the Internet. And the way around that, of course, is to
> not announce the CIX route at all, just as you are doing today.
No, your PRDB ends up overriding the CIX routing. That's a result of bad
configuration for those routes which are commercial, or shortcomings in the
routing protocols. In fact, our routing to Netcom goes through FIX-East
right now, but it <should not> -- both MCSNet and Netcom are CIX members,
and thus we <should> route through the CIX.
I note that Netcom also has an attachment to ANS. Ok, but does that mean
that the FIX should be used for transit from us to them? No. That's a
<commercial to commercial> link, and should not go via FIX-east. Since
MERIT and ANS control that routing database and announcements which cause
those packets to go through FIX-east, however, the fact that it <does> go
through there is your responsibility, not mine. In point of fact, some of
my other network numbers which are not announced to the NSF core <do> go
through the CIX to get to Netcom..... so the route announcements vis.a.vis
the CIX aren't broken -- the PRDB and the way it all goes together is where
the problem lies.
My argument is that to bias (or exploit an existing bias in) the route tables
in such a way that more routes go through you and then claim that "ANS talks
to more people than anyone else" is misleading at best. I bet that I could
rig things to be exactly the opposite if I wanted to (and had the access to
the tables required to do so).
It is not MCSNet's problem that your routing capabilities, never mind the
rest of the Internet's, are incapable of properly weighting these issues to
resolve the routing collision and do the "right thing" in the face of this
kind of attachments.
Do we need a new routing protocol? Probably. BGP, from what I can see, is
not up to the task of sorting this out. So what else is new? Is ANS
working on a new RFC for an Internet routing protocol that takes care of
these things? As a public project?
> And therein lie two problems:
>
> 1. The current CIX policy deters growth in the commercial Internet
That doesn't follow at all. In fact, the CIX <made possible> what we've
done here. If I had to take 20 attachments to get routing to the regionals
what I've done here would have been cost-prohibitive -- thus, MCSNet would
not exist in its present form.
> 2. All CIX members can bypass the connectivity restrictions imposed
> by the CIX Association filtering gateway, by preferring the NSFNET
> Backbone Service path, EXCEPT FOR ANS.
I don't see how this follows either.
> It is indeed ironic, that the organization established to grow the
> commercial Internet and demonstrate that government funding is
> unnecessary, is inadvertently increasing its members' dependence on
> taxpayer subsidies.
Excuse me?
> That is, they don't try to "backdoor" the membership of the CIX. This
> is a policy choice on their part, and one that I happen to agree with
> fully.
>
> Karl, If you are insinuating that ANS has chosen to violate CIX
> policy, then you are dead wrong. As a CIX member we have signed a
> membership agreement which we abide by.
Why was I offered a gateway connection and told (verbally, natch) that the
CIX membership was considered optional by ANS, yet ANS knew I intended to
resell bandwidth in many forms? In fact, in the contract <absolutely nothing>
was or is mentioned about the CIX or membership in it, nor the fact that I
would not have routing to CIX members (unless they also had ANS attachments)
unless I joined the CIX. THAT document <was> given to me in writing.
Does ANS represent itself as the entire Internet, or that it has connectivity
to it all by itself? That "the CIX doesn't matter"?
> Prior to becoming a CIX
> member in November, we had interconnected with the CIX under a
> mutually signed Memorandum of Understanding. Hearsay on this list to
> the contrary, ANS fully abided by the terms of that MOU and its
> routing plan. (ref: ftp.ans.net:pub/info/historical/cix-mou-routing.ps).
> If you have solid evidence to the contrary, feel free to discuss this
> with the CIX Association.
>
> -Ittai
I'd like to ask you to respond to the point above; we negotiated with your
firm before deciding not to go with ANS. One of the factors in that
decision was what I perceived as a severe difference of opinion on just
what our responsibilities would be via-a-vis the CIX if we did take the
connection with ANS.
If you wish to argue that "it wasn't necessary" then I ask if you're
willing to fund, at your own expense, direct attachments to all regional
connectivity providers, and allow any "indirect" traffic (which was
specifically prohibited in the contract ANS sent me). You can start with
MCSNet; our Chicago AGS+ stands ready for your line to be dropped here.
I'll even buy the CSU/DSU and port on the 4-T card if you're paying for
the rest.
If that is, indeed, not the case, then I ask you how a regional can get
connectivity to other regionals and commercial national providers other
than by (1) working out 20 contracts, and (2) paying 20 vendors every
month. Without using the CIX, now, since you're postulating that the
CIX actually harms the cause of commercial connectivity.
I'd like to make a further proposal. If you think you have a "better
mousetrap", let's talk. I'm willing to put up the "MIX" (Midwest Internet
Exchange) and allow people to talk through it on a set of terms which are
mutually agreeable if that is what "fixes" the problems. Of course, it has
to fund itself. Oh, I believe that any such venture has to be
settlement-free, although I'm willing to negotiate almost anything
else.
Any takers? I would expect that a call from ANS would be the first one I
receive tomorrow. My voice number is below.
--
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.
