We currently have L2 DCI with routers running VRRP, vmotion and our storage mirrors all the VMs over the link to another DC so failover is quite fast & easy. However we'd like to explore other options for doing failover after a disaster / manually migrating VMs to another DC (because of all the problems related to L2 DCI...)
What's the consensus today, should we run BGP on the Linux/Windows servers to advertise loopback address so we could failover the single IP address? Services are for our enterprise's internal use so it doesn't matter that much how the address block is advertised towards the internet.
Or should we have different IP networks in each DC and then update the DNS records when the host migrates? In this case, should we have load balancers to do this? We have F5 BIG-IPs but I would like to explore the open source world too, can for example nginx/haproxy do this?
This is also a 'it depends' question, but how do you see the correct way to run two DCs and be able to fail VMs over to another either after a disaster or manually before maintenance? It's not a single issue I'm trying to fix rather than get ideas how to do this correctly in the future. Our softwares we run are not in containers, I guess that would open more opportunities for this.
Thanks!
No comments:
Post a Comment