BGP Anywhere - Global Redundancy

Thu Apr 7 01:41:50 UTC 2005

On Wed, 6 Apr 2005, Vandy Hamidi wrote:

>
> I definitely want 100% of traffic going towards the Primary Site during
> normal operation.
>
> LocalPref/MED can be controlled by community strings with my direct
> peers.  As you said, I'm paying them for the service, but how will the
> advertisement behave after it propagates to their upstream peers?  At
> that point AS Path should be the only determining factor, yes?

Nope.  You're at the mercy of whatever traffic engineering or 
local-preffing other networks decide to do, and you won't have any control 
over it.

> Are ISP to ISP transit routes manipulated at MED or LocalPref levels?  I
> suppose some ISPs may mark some peer with a preferential MED.

Yes.

> I was turned on to BGP anywhere when reading up on UltraDNS.  Looks like
> they use it for Global load balancing in which a DNS server on the East
> Coast will respond to DNS queries to my East Coast DC and the same for
> the west coast.  They guarantee 100% DNS response, so I imagine it works
> for them.
>
> Has anyone on the list performed BGP Anywhere?  There has to be someone
> on Nanog that has done this.

This is more often known as Anycast.

I run the network infrastructure for the PCH Anycast DNS network.

It works well for trying to get traffic to come into multiple places. 
When we have a site go down, we withdraw the routing announcements from 
that location.

Trying to get traffic to go to only one place while sourcing BGP 
announcements from multiple places won't work very well.

-Steve