Getting a "portable" /19 or /20
jabley at automagic.org
Tue Apr 10 19:19:12 UTC 2001
On Tue, Apr 10, 2001 at 12:55:40PM -0500, John A. Tamplin wrote:
> On Tue, 10 Apr 2001, Roeland Meyer wrote:
> > AFAICT, multi-homing doesn't work unless you also have dynamic routing. Is
> > this wrong?
> Well, if all you care about is redundant routes (not optimizing paths) and
> don't worry about asymmetric paths, you can announce just your prefixes to
> each upstream and use only default routes (either received via BGP or
> static interface routes) for outbound traffic.
"announce" presupposes dynamic routing, doesn't it?
You can multi-home with everything routed statically, as long as you
have unsophisiticated requirements for choosing an outbound path, don't
need transport session stability, and can cope with some of your
addresses being unreachable from the rest of the network when one of
your providers stops being able to reach you.
Doing things this way can also make network problems harder to debug.
I multi-home like this at home, between a cheap DSL provider and a
cheap cable-modem supplier. So far there hasn't been a time when
both providers went down simultaneously. I have protected myself
from around forty-eight aggregate hours of downtime over the past
four months, and I haven't had to waste any time on the phone to the
cable operator helpdesk explaining to them the benefits of running
BGP to me.
I could have bought a pair of T1s to different providers and run
BGP; however, this would probably cost me around $3000 per month
more than I'm currently paying, would give me half the performance,
and would require me to buy a router with WAN interfaces (instead
of the OpenBSD box with 3 x $10 NICs).
Oh, and my multi-homing doesn't contribute to state bloat in the DFZ :)
More information about the NANOG