v6 addresses schemes have a very strong bias toward using the upper 64 bits for subnetting and the lower 64 for host identification out of the total 128. There are at least ways to pick the host part so far:
* use the EUI-64 mapping off a 48 bit ethernet MAC (flip the global/local bit, insert FFFE in the middle)
* get it from an upstream DHCPv6 server if the RA says to
* use a psuedo-random privacy address; windows OS likes this
* use a psuedo-random cryptographic hash address; the experimental SEND protocol would do this
* use a psuedo-random per subnet address, newfangled RFC-7217 style
* use a statically assigned address from the host administrator, using whatever takes his or her fancy
These are not mutually exclusive, e.g. you often have both EUI-64 mapped and privacy addresss valid simultaneously on a single interface on subnets where the router advertisements allow SLAAC.
It's common for server hosts with staticly assigned host parts to use simple things which reflect the service port or part of the v4 address; due to the multicast and global/local flag bits in the upper part the fun stuff is usually kept down in the low bits. So a complete v4 address might be embedded in the host part as 0:0:pq:rs. At the UW-Madison, the campus anycast recursive DNS servers have host parts ::53:1 and ::53:2, while the NTP servers have ::123:1, ::123:2, etc.
-- Jim Leinweber, WI State Lab of Hygiene
