North American Network Operators Group
Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: BGP Anywhere - Global Redundancy

  • From: Steve Gibbard
  • Date: Wed Apr 06 20:50:06 2005 
On Wed, 6 Apr 2005, Vandy Hamidi wrote:


All,
We're an ASP and are considering adding a secondary Backup Datacenter
(BDC) in the US to protect our web presence.

My goal is to ensure automatic failover of my Primary DC's (IP) traffic
to the BDC in the event of a catastrophic failure of the PDC.

I'm considering geographic load balancing and BGP Anywhere as the two
options.  I'm clear on how the Geo LB works, but have some doubts about
BGPAW as I've never implemented it before and documentation online is
pretty weak to non-existent.

Below is how I believe it should be done.

From PDC:

	-Advertise CIDR block to all peers w/good metric (0 hop count)

From BDC:

	-Advertise same CIDR block to all peers w/poor metric (+20 hop
count)
To clarify, you want no traffic coming into the backup site when the primary site is up, right?

Assuming a random set of peers and upstreams, this won't actually do what I think you're trying to do. Since local-preference overrides MEDs and AS path lengths, and since you don't have control over what goes on in other networks, you'll likely get some traffic coming into your backup site even when you don't intend it to.

You could *maybe* get around this by having the same transit provider (probably just one in this case, which is scary for other reasons) in both locations. If you're paying somebody money, you have a much better chance of getting them to follow your desired routing policy. Still, it's really not good to be making a routing announcement somewhere where you don't want to receive traffic.

You'd probably be better off looking into Cisco's "conditional routing" feature (I assume other vendors do something similar). This allows you to set a router to make an announcement only if it stops receiving some route, so you could have your backup site look for the primary site to go away and then start sourcing the route.

Failover time would probably be at most a minute or two, maybe better.

You could also look into various DNS-based ways of doing this.

-Steve