Re: IPv6 prefix lengths - how long?

[Mark Smith <[e-mail]>]

Hi Fred,

On Sat, 8 Jun 2019 at 03:44, Templin (US), Fred L
<[e-mail]> wrote:
>
> Hi, we all know about the tussle regarding the /64 boundary for IPv6 prefixes but
>
> assuming that we will one day want to allow longer prefixes the question is “how long”?
>
>
>

<snip>

>
>
> Because of RFC7934, we see the value of Host Address Availability Recommendations.
>
> It makes the point that, due to the nature of multi-addressing supported by IPv6, it
>
> would be useful for each node to be able to configure multiple (perhaps even many)
>
> IPv6 addresses from an IPv6 prefix. But, “how many”?
>
>
>
> My assertion is that a multi-addressed host (or, an End User Network router) should
>
> support numbering for at least 1K IPv6 addresses. This would imply that (assuming
>
> at some point the /64 boundary is relaxed) the longest IPv6 prefix should be a /118.
>
>

I think we really need to work out what a host is first.

Is it the physical device? Virtual hosts within a physical device?
Individual applications? Threads or processes within applications?


"On Many Addresses per Host", S. Bellovin
https://tools.ietf.org/html/rfc1681


"Transient Addressing for Related Processes: Improved Firewalling by
Using IPV6 and Multiple Addresses per Host", Peter M. Gleitz, Steven
M. Bellovin
https://www.cs.columbia.edu/~smb/papers/tarp/tarp.html


"node" is probably better to use to describe a network attached entity
that needs one or more addresses to be able to send or receive
packets.

The number of addresses a node needs depends on what it is, what it is
doing and how it is going to use its addresses. In some cases 1K would
be excessive for a node, for others it could be far too little.

As others have said, there can be other properties of addresses, such
as privacy, security, and plug-and-play operation. As an example, the
choice in 1980 of 48 bit Ethernet addresses, rather than just 10 bit
addresses as originally required to suit a maximum of 1024 nodes on a
link, has paid off immensely for nearly 40 years.

"Too many" addresses for a node, to the point where the unit of
allocation starts to cost something where the cost is worth worrying
about, is a much better answer than not enough.

I'd say that cost threshold is at /64 today (e.g. RFC 8273, "Unique
IPv6 Prefix per Host"), also placing value on it being the minimum
link allocation convention we've had for more than 20 years (since RFC
2373), and that there is more than approximately 3/4 of the total IPv6
address space unallocated for any purpose at all.


Regards,
Mark.










>
> Thoughts?
>
>
>
> Fred
>
> --------------------------------------------------------------------
> IETF IPv6 working group mailing list
> [e-mail]
> Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
> --------------------------------------------------------------------