[nsp] Methods for Non-BGP multihoming

Darren Bolding darren@bolding.org
Tue, 23 Jul 2002 13:37:46 -0700

Comments inline.


> -----Original Message-----
> From: cisco-nsp-admin@puck.nether.net
> [mailto:cisco-nsp-admin@puck.nether.net]On Behalf Of
> CARL.P.HIRSCH@sargentlundy.com
> Sent: Tuesday, July 23, 2002 12:36 PM
> To: cisco-nsp@puck.nether.net
> Subject: [nsp] Methods for Non-BGP multihoming
> Given the shakiness of WorldCom, I'm looking into
> implementing multihoming
> for our internet connection. Currently leaning towards a
> metro ethernet
> provider.
> I've read a couple of Avi Freedman's BGP tutorials as well as
> a number of
> documents from Cisco (BGP Case Studies, the ISPCon BGP presentation,
> Configurations for Load Sharing with BGP in Single and Multihomed
> Environments, etc). I feel like I've got a basic grasp of BGP but for
> simplicity's sake I'm still inclined to go with the option of
> taking static
> routes from each ISP and allowing them each to advertise our
> /16. I'm not
> interested in load-balancing, only redundancy.
> If both ISPs were to advertise our /16 address space
> (registered by us, not
> an ISP), we wouldn't necessarily need an AS number, would we?
> Given that we
> wouldn't be actually talking BGP with anybody and the ISPs would be
> handling the route advertisements.  We'd put a metric on our
> static routes
> outbound so only one link would be used unless it failed.
> Are there any serious gotchas to this sort of approach?
> Outbound routing
> would be very simple, but I'm wondering about the logistics
> of having ISP A
> and ISP B both advertising routes. When traffic comes inbound to our
> network, is there any way to make sure that one provider is
> always used?
> Does it even matter? It's the issue of how traffic finds its
> way to your
> network when you're multihomed that I'm not quite clear on.

It is generally preferred to avoid asymetric routes, as this can confuse
higher layer protocols (different delay on incomming and outgoing

One solution would be to have the backup provider pad their announcement
of your address (they would list their AS multiple times when announcing
the route).  Given that essentially everyone uses AS path length to make
routing decisions, this usually works- that is, everyone sends the
traffic to the upstream with the non-padded path.

I would reccomend padding the AS path at least four or five times.  I
would also reccomend asking your service provider if they provide this
service before you agree to a contract with them :^)

> If any body knows a better way to do this, I'd be happy to
> hear about it.

Plenty of options exist, especially if you are an outgoing only (all
traffic originates from inside your network) or if you are an
outgoing/small number of servers.

But given that you have a /16, I doubt that, and also you will likely
have little problem getting people to listen to the announcement for a

> I'm also curious about the pros and cons of running BGP with limited
> routing tables or even BGP with full routing tables. At this point the
> primary benefit of running full BGP appears to be improved
> visibility into
> traffic patterns and easier troubleshooting.

Well, you would also get to use both connections for traffic and the
likely improved performance of that.  At least in the case of full

> thanks much,
> -carl hirsch
> _______________________________________________
> cisco-nsp mailing list
> cisco-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/cisco-nsp

-- Darren Bolding       darren@bolding.org --
--                                         --