[nsp] inbound failover without BGP?

[Florian Weimer]

Sergio Ramos <sramos@gibnet.net> writes:

>> How do they control the inbound traffic?
>
>
> DNS.

So they don't control it, they just hope that they can change the DNS
as fast as possible. ;-)

> Any other idea on this scenario?

> That´s true. But are you sure that the connections would remain active using
> BGP?

At least over here, server links between IRC servers stay up when the
backup link has to be activated.  But that's probably just a fortunate
network topology.  (The network topology *does* change, it's not just
"we use link A instead of link B".)

> Let´s have a look at the BGP convergence in practice:
>
> http://www.renesys.com/projects/leiden/Labovitz-Leiden2000.pdf 

Ugh, these numbers are far worse than common wisdom would suggest.

> As I said I haven´t tested these solutions but I think that can be enough
> for many people that needs to be multihomed and even adds more control on
> the bandwidth than BGP.

It's certainly better than that what's happening here in Germany: I've
been told that cheap SDSL resellers start offering BGP peerings.  They
do not advertise it -- yet.

According to the rumors I've heard, their upstreams won't filter the
announcements (and the resellers themselves have hardly got enough
clue to keep BGP running at all).

Of course, these resellers offer much lower prices than the
established and small-scale ISPs who can help their customers with the
implementation of multi-homing scenarios.  If these rumors are true,
this phenomenon might be the end of the small-sized ISPs with clue in
Germany.

*sigh*

> I wanted to highlight that being multihomed does not mean to have your own
> IP addresses, AS number and increase the size of the BGP routing table.

Yes, a low TTL on the DNS record of your central mail relay can be
helpful, too (only if you can update the secondaries in a timely
manner, of course). :-)

I don't know much BGP (so I probably shouldn't post to this list ;-),
but the more I learn about it, the less confident I feel that purely
BGP-based multi-homing solutions are the solution to all multi-homing
problems.  (And BGP is indeed often overrated, as far as I can tell.)
So different views regarding multi-homing can be helpful, and I'm a
bit sorry about my initial reply now...

-- 
Florian Weimer 	                  Weimer@CERT.Uni-Stuttgart.DE
University of Stuttgart           http://CERT.Uni-Stuttgart.DE/people/fw/
RUS-CERT                          fax +49-711-685-5898