Re: [v6ops] Implementation Status of PREF64

[Mark Smith <[e-mail]>]

On Tue, 12 Oct 2021 at 19:13, Pascal Thubert (pthubert)
<pthubert=[e-mail]> wrote:
>
> When that’s the case, and the node needs, say, less than 1000 addresses, could we delegate a /118?
>
> This way, we can build a /64 NBMA subnet, which maintains the subnet boundary at /64, but allow routing inside.
>
> We could, e.g., terminate vxlan on the /118 and use routing over them.
>
> I would love to see that…
>

How many nodes are these NBMA subnets?

Why not give each node a /64? If you're doing flat routing, which it
sounds like you are, a /48 provides 64K /64s.

For GUA address space, APNIC RIR fees amount to about $0.03 Australian
dollars or less per annum (not per month!) for a /48 (and they're
effectively free if your IPv4 address space costs more than IPv6). GUA
/48 space is very, very cheap. (In any organisation paying RIR fees,
there probably isn't a cheaper per-unit cost resource. Pens in the
stationary cupboard cost far more than that.)

If you're using ULA address space, each ULA you generate is a /48, so
for example if your NBMA subnet is 250K nodes, you'll need only 4 x
ULA /48s to give each node a /64.

(With some of the EVPN work that is going on regarding ND proxying to
reduce multicast traffic, it has occurred to me that a lot of
organisations could do a single EVPN across their whole network,
fitting all devices within a single /64. In the 90s, before layer 3
switches, there was a saying - "bridge where you can, route where you
must". A big flat layer 2 domain, if you can do it, is quite a lot
easier. If there is any need for subnet boundaries, it's more likely
to be able to create security domains, although I don't entirely agree
with that, I think zero trust networks is a far better idea.)

Regards,
Mark.

>
>
> Pascal
>
>
>
> From: v6ops <[e-mail]> On Behalf Of Lorenzo Colitti
> Sent: mardi 12 octobre 2021 6:39
> To: Ted Lemon <[e-mail]>
> Cc: v6ops list <[e-mail]>
> Subject: Re: [v6ops] Implementation Status of PREF64
>
>
>
> On Tue, Oct 12, 2021 at 1:14 AM Ted Lemon <[e-mail]> wrote:
>
> However, they’re not going to want IA_PD addresses on the directly attached interface. They’re going to want IA_NA there and to route the IA_PD prefix to the host via ND resolution of the MAC of it’s IA_NA address.
>
>
>
> This isn't really necessary, although it's common practice. If the device is not going to use the prefix on the DHCP-ward link, it can use the address on the link it /does/ use the prefix on as its GUA. Its link-local address on the DHCP-ward link will suffice for routing, and is known to the DHCP server because that's the address it used to do the IA_PD.
>
>
>
> +1, that's how I think such an implementation would behave.
>
> _______________________________________________
> v6ops mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/v6ops