[nsp] inbound failover without BGP?

[Furnish, Trever G]

> -----Original Message-----
> From: Tony Tauber [mailto:ttauber@genuity.net]
> Sent: Tuesday, December 17, 2002 10:34 AM
> To: Furnish, Trever G
> Cc: Cisco NSP List (E-mail)
> Subject: Re: [nsp] inbound failover without BGP?
> 
> On Tue, 17 Dec 2002, Furnish, Trever G wrote:
> > Is there any way to provide failover routing of INBOUND traffic
> > using links from multiple providers, OTHER THAN running BGP with
> > those providers?
> 
> <snip>
> 
> > But wasteful or not, having inbound failover is a requirement for
> > most commercial web sites that provide service to paying customers.
> > If the link from Provider1 fails, the inbound connections still
> > succeed because Provider2 is still announcing to the world that he
> > has a path to me.  Is there some other way to do it?
> >
> 
> If you were to get physically diverse circuits from one provider that
> landed on different routers at their POP (these may be big "if"s), you
> would appear to have addressed the problem scenarios I've described.
> [...]

True, this addresses many points of failure and may be what I end up doing,
but it doesn't match *all* the resiliency available using BGP with seperate
providers.  If a provider has routing instability, using BGP I can just stop
my announcements to that provider and incoming traffic stabilizes.  If they
go bankrupt or lose their peering agreements with other tier1's I'm not SOL.

> While going to two different providers may get other-end router
> diversity it may not get physical diversity out of your building or up
> to the telco CO.  Do you have CPE-router diversity on your end to
> prevent router problems from causing outages? [...]

Yes, cpe and physical circuit paths are both redundant.

> Just asking because there seems to be unreasonable FUD in this area.

I get your point and don't disagree, and the comments are of course
appreciated, but I was hoping there was something new - a feature of some
routing protocol I didn't know about - that I'd slept through, that would
match bgp-multihoming's redundancy without matching its requirement for use
of a /24.

-t.