Re: sub-basement multihoming (Re: Verio Peering Question)

[kevin graham <kgraham () dotnetdotcom org>]

> I've noticed it, too... in some ways demand is even greater than
> among small ISPs who have an inkling about how BGP works.
>
> > Is my estimation that for at least some broadband providers,
> > per-household/per-customer BGP is a operational expense
>
> There are parties who are taking this into consideration.
>
> > capital purchase of new equipment, completely out-to-lunch (in
> > advance of an interesting new product launch in the next few
> > days)?

> Re the "high cost" of multihoming... perhaps now.  Most "smaller
> places" can't afford to multihome given the current cost of two
> T1s (hard to get BGP over broadband) and a Cisco that holds 128M
> (even "smaller places" seem to concerned about brand recognition,
> and are often reluctant to run Zebra).

Without trying to start a flamewar, this is one of the places where NAT is
exceptionally valuable. Setting up "redundant" connectivity for these
users given a set of n consumer-grade, commodity connections (DSL, dialup,
cable, etc) is rather trivial, and NAPT implementations have gotten robust
enough to accomodate most of the common layer-ignorant protocols. This
user doesn't want global route visibility, nor do they give a shit about
filtering or allocation policies -- they want to be able to get to their
pr0n when The Internet is broke.

This `knowledgable' SOHO user is most likely already using NAT to get
their office buddies online across the $40/mo DSL link -- likewise, the
use of NAT for multihoming isn't introducing any new complications into
their End-to-End Experience (tm).

For inbound services, where address distribution and portability is of the
most concern, SMTP is the most to worry about, and multiple equal-weight
MX's (to each of the PA addresses) take care of the problem. For a
business considering this, and interested in external services, if they
haven't already signed up for the $20/mo Web Hosting account from the back
pages of $PC_USER_RAG, convincing them to do so shouldn't be hard.

Given these caveats, the one problem that consistently comes up is
link-state inspection. For the low-end, people are using the ethernet -
<DOCIS,ADSL,2-way Satellite> bridge they got with the connection. PPPoE
makes this easier, as there's an interface on the router that will change
state, but otherwise its a guessing game. This is where providers aren't
even begining to play -- forget BGP peers to end-users, has anyone had any
luck getting any of the consumer-grade 'broadband' providers to do so much
as a RIPv1 default advertisment?

Of course, while this does keep the global routing table free of
non-aggregated micro-allocations, it does increase overall utilization.
Cleaner solutions for effective multihoming in the future are certainly
needed, but Multihoming For Dummies is quite obtainable today.

> However, I've encountered [consulting] customers with multiple
> _dialup_ connections who want to know if they can just balance
> traffic across both.  I think that the demand is there -- current
> products just don't allow it.

It might be pricey for that application, but their 1750 can do it just
fine. The pre-packaged "Internet Router/Firewalls" flooding the market
from Linksys/Netgear/etc haven't caught on yet, but its just a matter of
time.

..kg.. <back to lurking>