I kind of know what I want to configure for the following scenario but thought I would get your views on this:
2 x sites - Topology the same at both sites
LAC --> LNS --> CORE
From the core we have the exit to the upstream ISP and this is where our eBGP resides (PE).....
We will also have redundant links from one core in one data centre to another core in another data centre.
We will have 3, large IP v4 prefixs that currently, we are unsure will be assigned to subscribers from which data centre!!!!
So, this leaves us in the position that we will be advertising the full networks ranges from each site. Obviously, with the X-Connects in place between the sites there is no issue with the upstream routing, however, if we were to lose the X-Connects this would create a problem as we would only ever have one route back to these networks, and that would either be site A or site B.
A way around this issue is to create 2 pots of addresses. 1 for Site A and 1 for Site B and advertise those to the correct upstream peers. This still, however, leaves an issues if instead of losing the X-Connects, we just lose an upstream peer and iBGP will re-route out of the opposing site where the routes will NOT exist.
We looked at 2 options:
Option 1: If that were to occur we would have to manually add the prefixs to the other site, temporarily.
Option 2: We advertise all routes out of each site and then remove the routes when a site fails
However, and here is the question.... I was thinking that we could probably control this dynamically with BGP preference values..... Pretty certain it will use the preferred route until it detects the route down and then use the backup.....
What are your thoughts on using the BGP preference to achieve the desired results?
You will only be able to control your egress traffic direction with bgp local preference not the return traffic you get from the connected upstreams. So this can control your outbound traffic to prefer going out to the ISP in either DC.
From your description I assume the upstream peers are different at the two DC sites. If that is the case you cannot control return traffic in BGP metrics from them.
What you could do is advertise the full large prefixes at both data centers.
Then also adverstise at each data center the longer mask that you have internally assigned to each side. These because the are a longer match will be preferred by an ISP that gets both routes and thus traffic will mostly go to the desired data center during normal operations.
During failure the one side longer mask prefix will be lost but the full subnet is still availalbe to the other side so return traffic can still come back in.
Perfect response..... that's one of the two options we were looking into... or very similar.... thank you for the advice....... brilliant.... 🙂
I agree with the route back. Because they are separate ISPs from each site, the route back was always going to be the issue. But I think, as you have mentioned Steve, that the best option will be to have the full routes on each site advertised along with the smaller networks....... We can test before going live by pulling some connections (would much rather do this onsite than disable a port remotely and for some reason it doesnt come back up) 🙂