[clug] IPv6 for home (now aluminium foil hats, with an option on electrum and terracotta)

Francis James Whittle fj.whittle at gmail.com
Sun Jan 1 04:14:47 MST 2012


On Sun, 2012-01-01 at 19:19 +1100, Scott Ferguson wrote:
> On 01/01/12 18:15, Francis James Whittle wrote:
> > On Sun, 2012-01-01 at 16:25 +1100, Scott Ferguson wrote:
> <snipped>
> 
> > I use Avahi almost exclusively for local DNS.
> 
> If I could figure out how to make Avahi play nicely with DHCP so the PXE
> server still worked... I guess there's probably a "man" thingie I should
> squint at.

Your name resolution and IP address allocation should be independent...
It's not necessary to use "zeroconf" for allocation to use it for
resolution.  Blatantly stupid idea in IPv4, really, seeing as you can
only autoconfigure link-local addresses that way.

> 
> > It would be interesting to know if multicast DNS is any good at large
> > numbers (let's say 10000 or so) of hosts...
> 
> Or find a way of hardcoding the IP addresses - which is not hard, PXE
> boot, check for tag file, if not exist write /etc/network/interfaces ,
> assign incremented ip, add ip to dns table, write tag file, reboot.
> Either way it's unlikely to be something you can run on WRT (not enough
> RAM). But then there should be lots of alternatives to WRT in the next
> couple of years - re-purposed Kindles/FondleSlabs/Nokias..
> 
> > Sadly the only way I would have of testing this would be to boot up
> > a large number of virtual machines, and I'd probably run into memory 
> > usage problems even if I had a series of images that did nothing but 
> > boot up and launch an mdns responder....
> 
> Depends how many NICs per device - you could also (maybe) use timed
> changes of /etc/hosts and reset hostname - that way one device (P or V)
> could have many identities :-/

Not at the same time though.  We're looking for multiple concurrent
hostnames here.   I suppose it should theoretically be possible to have
multiple virtual interfaces with an mdns responder that responds with a
different name on each.  Abusive of the protocol and otherwise
pointless, of course....

> > 
> >>> For example, I could just plug up to around 2^64 devices into my
> >>>  network and... well I wouldn't be able to power them all, but
> >>> you see how that's 32 times the ENTIRE ipv4 address space,
> 
> According to the infallible Wikipedia:-
> "The standard size of a subnet in IPv6 is 264 addresses, the square of
> the size of the entire IPv4 address space"
> 
> So you're not exaggerating.
> 
> >>> and I still shouldn't have run in any major configuration 
> >>> problems (Except the problem of where to put so much stuff, and 
> >>> how to power it).  After that point I have another 255 networks 
> >>> of the same size to plug stuff into, because my ISP handed me an 
> >>> unreasonably large block of addresses.  Seriously, what could I 
> >>> possibly need a /56 subnet for?
> >> 
> >> Interesting - unfortunately my brain refuses to do big math 
> >> today.... like projected population size and uniquely identifiable 
> >> network connected devices.
> >> 
> >> Is it possible that the IPV6 address pool is large enough for 
> >> mandatory hard-coded addresses in devices instead of SA[*1]?
> >> 
> > 
> > I worry about the routability of mandatory hard coded addresses....
> 
> I'm sure those that propose tracking everyone everywhere have got a
> solution for that (10 minute ping times).

Considering the sparse nature of address pool population, I'm sure that
tracking everyone everywhere is still somewhat possible with SA....
 
> 
> A future scenario might be that devices are designed to be inactive
> unless activated by an authorised system at POS - after suitable ID and
> authorisation has been produced. So IPs are linked to owner with the
> ability for registrants (companies) to transfer the ownership, and a
> central authority to revoke them (deactivating the networking ability).
> With that scenario routablity would be considered irrelevant.

Sounds like the Antithesis of the Internet, that.  Don't want to think
about it....  One of the major points is that it's so decentralised it's
unimaginable to shut down.

> > 
> >> Perhaps I just haven't considered the number of possible discrete 
> >> networked devices of the future...?
> >> 
> > 
> > It's physically improbable to ever have that many things.  I'm 
> > talking about individuals filling up their /64.
> 
> Now that my hangover fades I don't think it is physically improbable in
> the longer term. If you had a lot of disposable devices with hardcoded
> IPs (in a similar manner to MACs). eg. grain of rice sized RFID devices
> for military,security, clothing,food, medical devices etc. Some of those
> could wind up occupying addresses on your network (in which case you
> could substitute "you" for "home").
> 
> Filling the device/node side could be your (apparently) smart meter,
> your vehicles, your portable devices, and anything you purchase that is
> network-able. Sounds implausible until I consider how likely a wireless
> gateway in most homes might be - which then increases the number of
> cheap RFID tagged devices that could be brought into the home network.

Let's do some maths here.  Start by assuming everyone lives for 100
years.  A reasonably generous average, even in the western world.
Exactly 100 years.  36,525 days.
Now let's assume that everyone "consumes" (for a wide reaching value of
consumes), on average, 20 things a day.  Another generous average, this
time overly.  In their lifetime, that's 730,500 things.
Now, let's say there's an average 6 or so people to a subscriber service
that has a /64 prefix allocated.
That's 4,383,000 things in a /64 prefix.  These things are not all there
at the same time, new things arrive, old things break, people come and
go, taking their things with them.
Each /64 prefix can handle, simultaneously, (2^64 - 2) unique IPv6
addresses ... That's in the order of 1.845e19; 4.209e12 (~ 2^41.937 or
so) addresses _per thing_.  And I'm effectively including clothes and
food here.

We really are talking about an outrageously humongous address pool here.

> 
> We've already got IBM building systems to track food from farm to
> checkout - the consumer gets that information real-time, the rest of the
> interested data users have to wait.
> How long before some genius decides your shopping preferences, eating
> habits, garbage statistics, etc is marketable.
> 

This is already marketable information.

> 
> Anyway, this aluminium hat is getting hot, and I really don't want to
> test it's ability to attract lightning during this storm :-)
> 

Avoid crinkling — smooth the surface of the hat.  Also helps against
microwave radiation.  Shaping it into a sombrero will deflect almost all
of the electrical force of any lightning strikes away from you.  The
more sombreros the better, but you should make the bottom one out of
non-conductive ceramic.

Cheers



More information about the linux mailing list