Date: Fri, 10 May 2002 09:48:25 -0400
From: James Smith
I think we all assume that our provider "guarantees" us some
sort of "total reachability". Near as I can figure, they do
not. Therefore, you buy a pipe into their network based on
percieved and actual connectivity and hope that the situation
remains static at best. Does ANY provider give a
"reachability" guarantee?
<iirc memory="bad">
Wasn't there a small russian ISP that had no access to _1_ during
the mid- or late-90s?
And didn't some ugly peering battles between 701 and 3561 back
when 3561 was MCI cause some { severely hampered | loss of }
connectivity between the two?
</iirc>
Help me out... I wasn't following routing and such very closely
back then. But it seems that none of this is new, just another
iteration of the same...
And didn't some ugly peering battles between 701 and 3561 back
when 3561 was MCI cause some { severely hampered | loss of }
connectivity between the two?
When AS3561 started (registered in 1994, turned on in 1995),
it started with many of the old NSF regionals attached to it.
This included what would become AS1. So, I'm fairly sure that AS701
didn't have anything against peering with AS3561.
It may be that you are remembering miserable performance. mae-east
had to mature in a hurry, mae-west, pennsauken, chicago were just
getting started. The CIX SMDS was used in 1995. I don't know when
the expression "SMDS = switching makes data slow" started being used
(possibly started out east with PSI <--> UUnet) but it is certainly
true that the first gen cisco smds and atm cards were miserable under
load.
IIRC, AS701 and AS3561 had the earliest and richest set of private
interconnects. Of course, "earliest and richest" is fairly vague 
-mark
James Smith wrote (on May 10):
Maybe it is possible he made a business decision based on the long term
costs involved with multihoming/redundancy vs. the loss of near total
reachability. He may have come to the conclusion that the probability of
that scenario occuring was not sufficient reason to multihome. His call.
It's worth pointing out it's not always a technical decision. Partcularly
when things are tight, the bean-counters and other senior management tend
to shy away from "redunancy" and "resilience" often in favour of "insurance
policies" and "controlled risk".
Similar business-decisions are what cause those networks to not peer.
Whether fair or not doesn't matter. Big companies are big businesses. Big
businesses like to remain big. They all have debt and thus need revenue.
A common view is that a peer is the loss of a potential customer. Drop
all your peers, gain some potential customers. (Sprint said this to me in
those words once)
While nobody has tried to take a "Tier-1" to court for what could be taken
as anti-competitive actions said providers will carry on - it's win-win
for them. The marginal loss of connectivity to *your* network is so small
from their perspective, there's no issue. If mutual customers complain,
they blame you for not connecting to them (from experience, and having
seen this done in black and white). The words used are along the lines
of "that is what happens when you connect to a non-tier-1, like us".
Just for reference, the European Peering Policy for one of the previously
mentioned carriers in this thread requires the announcement of 900+ /19's
from seperate LIR assignments, as well as the usual N-points connected,
M-bps transfered etc requirements. I'm under NDA so can't say more.
needless to say, we don't peer with them, and I don't buy transit from
them either, on principle.
We calculated that at the time only 5 IP providers in Europe (that were not
US owned networks) would meet that 900+ /19's requirement.
Chris.
Date: Fri, 10 May 2002 18:29:23 +0100
From: Chrisy Luke
[ snipped ]
While nobody has tried to take a "Tier-1" to court for what
could be taken as anti-competitive actions said providers
will carry on - it's win-win for them. The marginal loss of
connectivity to *your* network is so small from their
perspective, there's no issue. If mutual customers complain,
they blame you for not connecting to them (from experience,
and having seen this done in black and white). The words used
are along the lines of "that is what happens when you connect
to a non-tier-1, like us".
Now, as much as I'd not expect C&W to peer with us, look at
PSINet. Were they small? What about EXDS? Those peering paths
were to provide better-<insert various metrics> to the eyeballs.
I'd argue that both are/were significant. And as much as it's a
good thing to not require everyone to peer with everyone (n^2
would be out of control), it would also be bad if the entire
world depended on a single ASN.
I agree that a line must be drawn, but disagree with where
certain carriers draw the line. But I suppose that we're
insignificant to them, and they probably don't even care about
selling _transit_ to someone so small. [Not that this is
inherently bad... just be up front about it like L3, and tell
people what the minimum is.]
I guess the C&W slogan is also rubbing me the wrong way.
"Delivering on the Internet promise" seems to imply that traffic
gets there reliably. 
[Note that I'm impressed with the good
community support... not just bashing C&W.]
Note that this is not peculiar to the Internet. Look at the EDI
world, and what happened to ICC with Sterling and GE. _That_,
IMHO, is a much more clear-cut case of anti-competitive behavior.