Re: [**EXTERNAL**] Re: the race to the bottom problem

[Philip Homburg <[e-mail]>]

>However, the privacy arguments are fairly strong for not going beyon
>d /80, I think, at any time in the future.

Peronally I'm a lot more worried about collisions. At the moment we have
no operational experience with collisions. Does DAD work? We don't really
know. How does a device recover if it picked an address that is a duplicate
of another.

If we look at /96. That means that more than one in 4 billion addresses
generated is a collision. If IPv6 is adopted worldwide, that will be many
collisions per day. 

A key question here is: do we set a relatively safe limit that allows us
to ignore DAD because duplicates essentially never happen, or do we let 
operator pick a prefix length, spend quite a bit of time trying to deal
with all the collisions that result.

For example, at the moment a low power IoT device could just switch off for
a while and resume. Technically that would be wrong, but in practice it works.
What would we do if IoT devices would randomly renumber do to collisions?

If we set a hard limit, then we are back at the original question, how do you
extend network if you got the longest possible prefix that is supported.