Re: draft-bourbaki-6man-classless-ipv6-00

[Lorenzo Colitti <[e-mail]>]

On Fri, Jun 2, 2017 at 11:56 PM, Job Snijders <[e-mail]> wrote:

> >    - The document lists no compelling use cases of what you can do
> >    with less than 2^64 addresses per link than you can't do with a
> >    2^64 addresses per link.
>
> This is not true. The security section for instance shows: "In such
> cases, the use of smaller subnets forces an operational limit on such
> data structures, thus helping mitigate some pathological behaviors (such
> as Neighbor Cache Exhaustion attacks)."
>

I said "compelling". Those attacks have relatively simple operational
countermeasures. They were discovered several years ago, and we're coping
just fine.


>
> >    - The only technical motivation in this draft seems to be "classful
> >    addressing was a bad idea in IPv4". That's not a valid argument in
> IPv6
> >    because the address space is completely different. A /64 is *four
> billion
> >    times* bigger than the IPv4 internet. That's 10,000 times more than
> the
> >    difference between a grain of sand and the whole planet we live on.
> Such a
> >    huge scaling difference pretty much invalidates any argument that
> solutions
> >    that worked well in IPv4 will work well in IPv6.
>
> Would be a shame to throw away lessons learned from IPv4.
>

I think that's exactly the fallacy here. It is not useful to apply
something you learned about IPv4 addressing to IPv6, because the number of
addresses is incomparably bigger. Rule of thumb in software engineering is
that a solution usually scales by one or two orders of magnitude. Here we
have something like 34 orders of magnitude. Even a single /64 is 9 orders
of magnitude larger than the IPv4 Internet. It really doesn't make sense to
assign addresses the same way.


> Making IPv6 look more like IPv4 (for instance through broader adoption
> of DHCPv6 w/ routing options, and classlessness like in IPv4) will
> positively impact IPv6 deployment.
>

But the deployment we get will be dumbed down to the level of functionality
that we can support (and suffer from) in IPv4. It's not worth it.

Example: I don't want us to have to deal with NAT any more, ever. If you
think we can use DHCPv6 without ending up with NAT, then please read
RFC7934 in a loop until you understand why we can't. :-)