routing around Sprint's depeering damage
mpetach at netflight.com
Sun Nov 2 19:30:15 UTC 2008
On 11/2/08, Adam Rothschild <asr+nanog at latency.net> wrote:
> On 2008-11-02-10:14:14, Matthew Kaufman <matthew at eeph.com> wrote:
> > But seriously, it shouldn't be necessary to have two connections at
> > work [...]
> This is less than clear, and largely dependent on a specific
> organization's [in]ability to function if their internets go down.
> End-site multihoming in some form or fashion is a growing requirement,
> and folk thinking otherwise need to get their heads out of sand.
> If anything, these recent de-peerings underscore the lack of wisdom in
> end users connecting to (or purchasing CDN services from) members
> of the tier 1 club directly.
Thank goodness IPv6 cleanly supports end-site multihoming so we
won't ever face messy issues like this in the Internet of tomorrow!
Oh, wait--could this end up being a damper on IPv6 deployment?
"I'd like to move to IPv6, but I can't multihome in IPv6, and I've seen
what happens when you don't multihome--so I'll stick with v4, where
I at least have the option to multihome to try to avoid being screwed
when the net is partitioned like this."
Hopefully people recognize that we're rapidly being caught between
the twin perils of Scylla and Charybdis here; on the one hand, we
don't want to mandate universal connectivity throughout the Internet,
we want to allow networks to engage in squabbles like this, and we
tell companies "hey, this is the reality of the internet--you want your
customers to have more reliable connectivity, you need to multihome"
But at the same time, we're telling them "IPv4 is running out, you need
to look at moving to IPv6; oh, by the way, in IPv6, you don't get to
multihome, you get your addresses from your upstream, and you're
stuck with them; you can buy from multiple upstreams, but you'll
have to use some type of kludge to switch addresses to make use
of the additional paths."
With network partitioning becoming more and more an accepted
fact of the Internet, if multihoming in IPv6 is not made at least as
easy as it is in IPv4, companies who cannot get PI space will not
move to IPv6 for any serious production traffic; they have heard us
chant the "you must multihome in order to reach the entire Internet,
partitions happen on a regular basis, and we refuse to let anyone
put regulations in place to prevent them" mantra enough times to
realize that the only viable business model for the forseeable
future is to use IPv4 addresses in an end-site multihomed fashion.
This is the bed we have created for ourselves; why do we spend
so much time here wailing and wishing it were otherwise?
More information about the NANOG