How is the IPv6 network prefix hierarchically allocated from registry down to subnet?
Allocation flows downward: a Regional Internet Registry hands a block to an ISP, the ISP hands a /48 "site" to a customer, who carves it into /64 subnets.
The prefix is built in nested aggregation levels so routing tables stay small:
| Level | Typical prefix | Who holds it |
|---|---|---|
| RIR (Registry) | ~/12 | Regional Internet Registry |
| ISP | ~/32 | Internet Service Provider |
| Site | /48 | the customer / company |
| Subnet | /64 | a single LAN segment |
The older textbook names for these aggregation fields are TLA (Top Level Aggregator = super-provider), NLA (Next Level Aggregator = provider/large enterprise), and SLA (Site Level Aggregator = subnetworks). For unicast addresses the leading prefix bits p were originally 001.
This hierarchical structure means a backbone router only needs one route entry for an entire ISP, not one per customer — that's route aggregation, and it keeps the global routing table manageable.
Tip: The exact field layout changed several times during IPv6's history; what stays true is the principle — provider-based aggregation so the internet's core routing tables don't explode.