Re: [v6ops] Implementation Status of PREF64

[David Farmer <[e-mail]>]

On Fri, Oct 1, 2021 at 12:50 AM Lorenzo Colitti <lorenzo=
[e-mail]> wrote:

> On Fri, Oct 1, 2021 at 2:15 PM David Farmer <farmer=
> [e-mail]> wrote:
>
>> So, is there a compromise somewhere between 1 and 2^64 addresses to be
>> had? Isn't it possible to request multiple addresses, that is to request
>> multiple IA_NA options? Then if the server returns a number of addresses
>> below a threshold, then refuse the addresses?
>>
>
> Well, let's find out. :-)
>
> Let's see if we get enough consensus to work on an RFC that says that all
> general-purpose DHCPv6 clients are entitled to receive N addresses; that
> any network that does not provide that number of addresses is in violation
> of the RFC; and that any client disabling the IPv4 stack if it does not get
> at least N addresses is in full compliance with all IETF specifications
> including RFC 8415.
>
> Personally I will be very surprised if you get consensus on any N even on
> this thread (let alone the WG or the IETF). The reason for that 1 is
> unacceptable to some folks, but any time you go beyond 1, then lots of
> infrastructure needs to be changed. And that's the rub here: IMO, when
> folks say they want DHCPv6, it's not because they want security or access
> control or tracking or whatever else. It's because they want to run their
> networks *in the same way as they do in IPv4*. That means an IPAM that
> maps one MAC to one IP and one DNS name, for example. If you're an
> enterprise, that's a legitimate desire: why pay costs that you don't think
> are necessary? The problem is that the enterprise does not pay the costs of
> NAT; device manufacturers, app developers, and users do. It would be good
> for users to stop having to pay those costs when we upgrade to IPv6. This
> is where we're stuck I think.
>
> Another reason for not having a hard number is that TCAM is expensive and
> that any number that would provide enough flexibiilty in the long term
> (e.g., one IPv6 address per app on the device) would likely be too
> expensive today. That's obviously easily solved with DHCPv6 PD, but the
> problem with DHCPv6 PD is that it's not IA_NA, and requires infrastructure
> to be changed. Same problem as above.
>
> Anyway, we can try this out as a thought experiment. I'll start: how about
> N=64?
>

Well, we run a quite sizable IPv6 network, I haven't looked at the numbers
in a couple of years, but we were seeming an average of 4 and change IPv6
address per MAC address including LL, we had some outliers with a few
hundred and even one with more than a thousand. Let's be generous, call it
5 IPv6 GUA addresses per host on average. I think almost 13 times average
might be a bit much, but 2 to 5 times the average seems reasonable, so I'd
probably start with a limit of 25 per host on our network. But, I need to
go look at what the 90th, 95th, and 99th percentiles are at, I just don't
remember.

So, for an RFC how about 10, I have a much bigger budget than a lot of
networks.

Next bid, please.

Thanks
-- 
===============================================
David Farmer               Email:[e-mail]
Networking & Telecommunication Services
Office of Information Technology
University of Minnesota
2218 University Ave SE        Phone: [phone]
Minneapolis, MN 55414-3029   Cell: 612-812-9952
===============================================