a radical proposal (Re: protocols that don't meet the need...)

Bill Stewart nonobvious at gmail.com
Tue Feb 21 21:03:50 UTC 2006

I looked at some of these models back in ~2000, but the dotcom boom
ended and I didn't get laid off from my day job, so I didn't go
trolling for venture capitalists, and my employer sold off their cable
companies - since then, the market economics have changed a lot, and
routers have started to support enough memory to keep up with the
demand.  The big questions about the dual-homed customer base are what
kind of connectivity they really need - Primary/Backup, or Primary /
Backup+extrabandwidth, or truly load-shared, and also what diverse
topology is available at the bandwidth they need.  For a reasonably
large chunk of the ~Y2K market, the answer was "A T1 or two with
cable-modem backup", and another chunk was "T3 or bigger, able to
afford a telco or CLEC access ring", and most customers were more
concerned about backhoe fade, which takes a long time to fix, than
about ISP routing glitches, which were less common than 5 years
earlier and usually had a much shorter mean time to repair.

None of these solutions requires a World Domination Grand Master Plan
agreed to buy everybody before it can be deployed - almost anything
can start out with two carriers or a transit-buying service provider
and then grow.

One obvious business model to serve the smaller market was to start a
"Slash-19.net", which would get a routable chunk of address space, buy
transit from one or two colo providers, and use GRE/IPSEC/L2TPv3
tunnels to connect to the customer through whatever Layer 3 media is
available, e.g. cable modems, and optionally use LEC frame or similar
transport where available.  In the emerging IPv6 world, a tunnel
broker service could do something like this.  And for equipment-cost
reasons, you'd probably use PCs instead of routers as your tunnel

Another business model would be for a Tier 1 or Tier 2 ISP to do
something similar, using a smaller chunk of their own address space,
and using a tunnel server at one of their peering points (or colo
space served by another ISP) to handle tunnels through the secondary
carrier, such as cable modem companies.   Making the addresses work
well would require them to use the dual-homing address space for those
customers' interfaces instead of whatever probably-geographical schema
they use for single-homed customers.  The cable companies would be an
obvious ISP to do this - they've got control over the most common
small diverse access methods, and most of them use PPPoE to connect to
their customers  so they've already got tunnelling.  New wireless
access ISPs could do much of the same business.

Another model is cooperation between big carriers - if you're doing
the N**2 pairs-of-carriers model, there are ~30-35 Tier 1 carriers in
the US, so ~1000 address blocks would be enough (if it sounds like a
cabal, too bad), and probably a similar but smaller number for Europe.
 Tier 2 players might need to arrange separate deals with one or more
of their upstream Tier1s, so they might double their address space
(still only adds ~10K routes), or else they might do an exchange point
approach (e.g. somebody like Linx starts Diverse-Linx.)  If somebody
can get more than two Tier 1s to cooperate, they could do the
geographic approach, which can make a major dent with ~50-100 cities
in their market.

More information about the NANOG mailing list