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

[Job Snijders <[e-mail]>]

On Thu, Jun 15, 2017 at 01:04:18PM -0400, Warren Kumari wrote:
> On Thu, Jun 15, 2017 at 5:20 AM, Philip Homburg <[e-mail]> wrote:
> >> SLAAC doesn't work unless the network provides enough address space
> >> to make collisions vanishingly unlikely.
> >
> > Indeed. So any RFC that makes smaller pseudo random IIDs possible
> > has to have at least a section that analyses collision
> > probabilities.
> 
> So, a couple of years ago I was doing some work with Juan Carlos and
> Dan Harkins in the IEEE on MAC Randomization (basically randomizing
> your (WiFi) MAC address to improve privacy / thwart pervasive
> monitors).
> 
> There was some discussion on how many bits you would need to randomize
> to make it unlikely that multiple machines will independently choose
> the same address. I argued that if you randomize 24 bits, and have a
> stadium of ~2000 people, you will basically never get a collision.
> Dan disagreed, and so, in a fit of pique, I wrote an App Engine app to
> prove him wrong -- and achieved exactly the opposite. This is a
> birthday paradox problem, and you will get a collision once every ~9
> times.
> 
> I've just updated my little app and put it here:
> https://ipv6-collision-probability.appspot.com/
> 
> It's interesting to play with the numbers and see how likely a
> collision is, given a certain subnet length and number of hosts.

Hah, nice :)

> >  And set a lower limit that is generally safe.
> >
> > Just leaving it to an operator to figure out that a /112 prefix will
> > sometimes lead to an IoT device failing due to an address collision
> > is extremely poor protocol design.
> 
> Example: With a /112, and 100 hosts, you will have a collision every
> ~14 times...

Luckily there are other methods of numbering things: manually, or
programmatically, or through DHCPv6. I believe there are more use cases
then IoT devices. SLAAC is not the only way of doing things.

Kind regards,

Job