Cisco Support Community
Showing results for 
Search instead for 
Did you mean: 
Community Member

Purposefull Abuse of OSPF

I'm working on setting up a fully redundant / load balanced dual site server farm. The two sites have an OSPF operated mesh going between them, giving multiple paths for ip connectivity. In each site, I'd like to hang a pair of ip networks, which will be the service ip's for my different services. These service ip's will live on load balancers to spread the load across the servers in each site.

Now for the odd part. Idealy I'd like each service, like say, pop, to have one ip, that loadbalances between the two sites and across the equipment without having to do any L2 bridging. Unfortunatly I don't know how to do that. So, my second best plan is for two ip's per service, each in different subnets. The two subnets represent the two sites. Subnet A for Site A, Subnet B for Site B, etc. This lets DNS do crude balancing between sites, and the loadbalancers handle inter-site availability. If I loose a site, half of all connection attempts (those to the second ip) fail though. As a hack/work around, I'd like each site to also advertise the OTHER site's subnet, only with a gigantic metric.

Site A:

Subnet A, metric 1

Subnet B, metric 255

Site B:

Subnet A, metric 255

Subnet B, metric 1

The loadbalancers would be configured with service ip's in both subnets. If my understanding of OSPF is correct, when both sites are up, traffic will be routed using the lowest metric. So site A gets all Subnet A traffic, site B gets all Subnet B traffic. If something happens so Site B falls off the network, the now huge metric announcement for Subnet B becomes the only one, and as such all subnet B traffic goes to site A.

Is this a viable configuration or is there a better solution?


Re: Purposefull Abuse of OSPF

If I understand what you are proposing to di, it's give both servers the same ip address, and then let OSPF route to the closest one. This is a common concept in ipv6, called "anycase," and there is a similar thing in multicast routing, when you have more than one RP on a PIM SM tree.

This shold be possible, and shouldn't cause any problems at all, if configured correctly.



CreatePlease to create content