Server addressing for renumbering ease

[George Bonser]

> 
> Why do you find it a bad idea for other servers? Just the switching
> NICs/mac address dependence issue? I'll grant that SLAAC is another
> moving part, but it's a pretty simple protocol that I'd be willing to
> trust with numbering my servers.

Ok, take a load balancer configuration where you have an IP that balances connections across several servers. The configuration of those servers on the load balancer is generally done by IP address but can be done using host names.  If you use host names, you have to depend on DNS being correct, the internal domain functioning properly, and the DNS server handling that data center to be working and reachable, and you have to depend on DHCP working when the server gets the address.  One failure scenario: imagine a power failure in an unattended colo.  The machines boot back up once power is restored but there is some issue that prevents the DHCP server from starting properly.  Maybe it lost a disk drive.  Now the service being provided to the Internet fails because none of the servers were able to obtain an IP address even though they are otherwise perfectly capable of operating.  Static addressing avoids that problem.

> Another thought I had is that with the rise of virtualization, SLAAC
> actually starts to make more sense for virtual machines. While
> replacing a NIC in a physical server isn't uncommon, since VMs have
> virtual mac addresses they can reasonably be expected to keep the same
> mac, and thus the sam SLAAC IP, for the lifetime of the VM.

Many operating systems will allow you to assign a MAC address to a physical interface, too. And there is "address space" for that region. Think of it like "unique local" MAC addressing.  That information can be stored in the configuration scripts run at interface configuration time and so the apparent MAC address can be static across hardware changes.