All nodes, workloads and services have an overlay IPv6 as well as multiple IPs.
The interface address is usually derived from SHA of the public key.
For network, options are to use the FC::/8 range (not defined, reserved for local), or FD::/8, which is specified as 48bit random and 16 bit 'subnet id'.
The 48bit can be derived from the SHA of the fabric public key, subnets may define different locations.
https://en.wikipedia.org/wiki/Reserved_IP_addresses https://datatracker.ietf.org/doc/html/rfc6890 Special-Purpose IP Address Registries https://datatracker.ietf.org/doc/html/rfc5735 - IPv4
0.0.0.0/8 - 'this network'. Can use 0.0.0.0 as source to 'learn the IP'. 10.0.0.0/8 172.16.0.0/12 192.0.2.0/24 - TEST-NET 198.51.100.0/24 - TEST-NET-2 203.0.113.0/24 - TEST-NET-3 192.168.0.0/16 198.18.0.0/15 - benchmark, not routed
100.64.0.0/10 - shared address space https://datatracker.ietf.org/doc/html/rfc6598
224.0.0.0/4 - multicast - a huge space that can be used ! 0xE0.00.00.00
240.0.0.0/4 - future use - 0xF0.00.00.00
169.254.0.0/16 - link local, auto-config (no DHCP allowed). rfc3927 169.254.169.254 used as MDS. Also DNS in GCP 'can't be used by any router' - but ok on the bridge 169.254.169.123 - NTP in AWS 169.254.169.253 - DNS in AWS 169.254.1.0 to 169.254.254.255 is the range for random addresses, 0 and 169.154.255.0 are free. recommends consistent random generator (MAC?), storing an assigned address, plus ARP probe and announce same domain with ARP announce - i.e. bridge
192.0.0.0/24 - IETF assignments 192.88.99.0/24 - 6to4