Entire ISP Forced to Close

Bill Stewart bill.stewart at pobox.com
Tue Jun 12 00:58:58 PDT 2001


At 11:50 AM 05/16/2001 -0700, Eric Cordian wrote:
>Jim Dixon wrote:
> >> Still, the Internet is for the most part a Star Network, with only the
> >> very largest providers multi-homed.
>
> > This is not true, unless your definition of 'the very largest' is very
> > loose indeed.  There are many thousands of multi-homed ISPs.  People
> > periodically attempt to draw graphs of the relationships between
> > ISPs.  If you look at these you see nothing similar to a star network.
>
>This hasn't been my experience here in the US.  I am familiar with about
>10 ISPs, from small mom and pop operations, to mid-size regional
>providers.
>
>The smallest ones have a single line.  Even a pretty big ISP can run on a
>single OC3, with a backup DS3.

Most of my experience is with big backbone providers,
big enterprises, web hosting services, and very small ISPs.

Early on, there were three main backbone providers - MCI, UUNET, Sprint -
and a small ISP would buy their first T1 feed from MCI (cheapest),
and as soon as they could justify a second T1, they'd buy it from
one of the other providers so that hopefully there wouldn't be
bad routing instabilities on both at the same time.
Things have gotten much more reliable, but also much bigger,
and most ISPs still buy diverse connections when they need more than one.

Almost every web hosting ad talks about having multiple connections,
whether that's 2 T3s or 2 OC12s to different backbones,
because you still need it for reliability.
If you're out of service for a day, you lose customers, fast,
while if your performance is doggy for a day, they'll usually stick around.
Having N thousand small ISPs, and hundreds of small web hosting businesses,
plus dozens of big ISPs and hosting services means there's lots of 
competition -
if you provide undependable service, people will leave, unless
they're somewhere geographically special or have other special issues.

>There are a few with a handful of OC12 and OC3 circuits, but these were
>generally obtained for specific customers.
>I can't imagine an ISP with 50+ distinct peers, with separate circuits
>to each.

Most non-huge US ISPs don't have large numbers of physical peering circuits,
but ISPs that use the public exchange points or carrier hotels
often peer with a number of other ISPs, because that either requires
just administrative agreements (on a routed exchange point)
or additional PVCs (on an ATM exchange point.)
Some exchange points work by everybody peering with the exchange
rather than with each other, but it's a similar effect.





More information about the cypherpunks-legacy mailing list