IPv6 Subnetting - Block RA on Switches?

Question

I'm currently planning a v6 only small datacenter network. Imagine for now one deployment consisting of 6 racks with 12 servers per rack. Each server has 2 NICs for two physically separate networks (management & data). The racks are interconnected.

For ULA addresses I want to divide 16 Subnet Bits with 4 bits for "network type", 4 bits for "deployment ID" and 8 bits for "rack ID". So each rack gets its own subnet and humans can identify them easily. So per rack I need to send RAs with the reachable on-link prefixes of the other racks and one prefix for SLAAC so the servers generate their addresses. But the racks don't have physical routers. Only the distribution layer above the racks has routers.

My idea now is to let one server send RAs with "Router Lifetime = 0" (which is allowed in that case (see RFC4861)) for the rack. But now the RAs spread over all racks.

Thesis: I need to block RAs from traversing through switches.

Question: Is this how it is meant to be done? Are there any operational experiences with datacenter networks on v6 only?

Answer 1

No, this is not how it is meant to be done. The normal way of addressing in IPv6 is to use one /64 per (V)LAN / ethernet segment and number the devices within that /64. So if the network stretches across multiple racks then don't put the rack number in the subnet id, but somewhere in the last 64 bits.

Answer 2

Without more details I might be generalizing here... but why don't you give each rack it's own VLAN which is trunked from the distribution layer?

Then have RAs sent for each rack from it's given Vlan SVI. Thereby ensuring you're not sending RAs left, right and centre. Each rack will be given its own subnet as it is in a separate L2 domain and it can configure its host bits via SLAAC.

Also the previous answer is correct regarding best practice... /64 for subnet... allow host to have 64 bits for SLAAC for eui-64 address configuration.