On Tue, Oct 1, 2013 at 11:11 AM, Elvis Velea <elvis@velea.eu> wrote:

Hi Jan,

Hello again, Elvis. :)

I'm not talking about how much you would use on your router or reverse dns.

I'm talking about how much to 'reserve' as minimum for a point-to-point link or for a service.

…

It's not preventing use of a smaller prefix, it's preventing assigning/sub-allocating less than a /64 for anything.

…

Well, I used to work as an IPRA at the RIPE NCC and my understanding of the policy then (and now) was that assignments and/or sub-allocations of anything below a /64 is out of scope and even if one IPv6 address is used within a /64, the whole subnet is considered to be used.

While this particular paragraph is fine, the others are less than, uhm, clear.

To me, this is the difference between letting me use e.g.
2a01:5b40::80:88:dead:beef:cafe as the IPv6 address for www.oyet.no
<http://www.oyet.no>, and having to use e.g. 2a01:5b40:88:cafe::1/64.

I don't actually see it like that. You can still use the whole IPv6 address to number a device, it's just that you can not split a /64 for different services.

For example, you can use a /64 to number, let's say, 100 devices that are in the same vlan doing the same thing and providing the same service but you can not number 100 different customers within a /64.

… because of this.

This doesn't make sense.

Why can I not number 100 (or indeed, hundreds of thousands of) addresses for different websites within a /64?

I think, perhaps, that the words "customer", "service" etc. are poorly defined and lead to significant misunderstanding, and that I have not made myself clear.

Whatever a hosting provider chooses to do with their IPv6 space to perform comparatively fine-grained routing and other network organization, should be pretty much irrelevant, as long as what's exposed to peers is sane.

The example I came with is not randomly chosen, I know there has been similar issues with the understanding of semantics in IPv4 policy, which according to some NCC members seemed to imply that it would be impossible to use a separate IPv4 address per SSL site without comparatively major bureaucracy, since (the argument went) these addresses should be assigned and registered.

Now you seem to be telling me that I cannot have a single webserver serving two different websites with two different IPv6 addresses, if they are both within the same /64.

If this is really the case, IPv6 is, within RIPE, a 64-bit address space, not a 128-bit address space, and depletion is guaranteed to become a problem with the proposed policy, as well as the current policy, considering that every hosting provider would need _many_ /32 allocations, as I understand the policy and the argument you make.

If, however, we consider IPv6 an 128-bit address space, and the rightmost /64 as simply not governed by RIPE policy, then a /48 is a _humongous_ address space, providing a LIR with 65k networks which may be redelegated to lower-level customers, such as e.g. web hosting providers, as long as these are routed as a /64 block.

--
Jan