[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: IPv6 adoption. 128 bit addresses are not enough!

> > This is going nowhere. Do the math and give us an even remotely
> > plausible
> > concrete scenario, in which we could run out of IPv6 addresses any time
> > soon. With the calculations, please.
> > 
> > Just out of curiosity, how many addresses would be enough, in your
> > opinion?
> 173 bits would be enough to adress every proton and electron on Earth
> 192 bits would be enough to adress every proton and electron in Solar System
> Even with nanotechnology, we will never run out of IPv6 adresses
> in predicable future.

enumeration <> addressing and routing

1) Contemplation about p+ and e- and enumeration of bits.

Lets assume you could enumarate most subatomic particles with 173 bits.
I guess you are opposed to nuclear power by your shunning of neutrons, but
thats off topic, even for this email.

Thats about 10^52 or so particles.  I will assume the average atom, crystal,
etc, will contain 10 addresses or 10 routable subnets, in a generally star
shaped network.  Obviously atoms with an atomic weight greater than boron or
so will need IBGP sessions between their multiple subnets to connect
internally and EBGP for interatomic communications, etc.

Longest traceroute would have 104 hops.

Anyway at any temperature above absolute zero the BGP sessions would quickly
go nuts trying to maintain an coherent routing picture.  Major route
flapping, I expect.

2) How to exhaust the 128 bit IPv6 routing space, very easily

Assign addresses geographically.

International tier networks for international routing will be 48 bits.  
To the nearest byte level boundary we need 16 bits for countries and 
16 bits for providers and 16 bits for the providers international routers.
If you have massively parallel routers a provider could have move than 
60000 or so routers, plus you have to allow for internal subneting. 

This allows nice easy access lists to ban specific providers and countries
from your internet space.

Then figure maybe 32 bits for regional ISPs to divide up their routers. 
Maybe 16 bits for different providers, maybe 16 bits for the routers 
themselves, probably subnetted.

Then figure maybe 32 bits for end user modem bank type things.  Get rid of
this dynamic PPP ip address garbage.  With all the appliances and stuff
having internet connectivity, you got at least one class C in every room of
the house.  Figure your average house would need one big (sparse) class B. 
Ease configuration, try "upper 112 bits".X.Y.10 (=kitchen).1 (=sink)

So our full IPV6 address map could be

Country (16) - 2^8 is too low, 2^24 is too high
International Provider ID (AS number?) (16) AS # = 16 bits
Subnetted international gateway routers (16) 254 gateway class Cs
Regional provider ID (16) AS # = 16 bits
Subnetted regional gateway routers (16) 254 class C's
Enduser provider ID (16) AS # = 16 bits
Enduser modem bank / whatever (16) Easily have 65000 lines in a central
Enduser (16) just to have a class B house with class C rooms.

Adds up to 128 bits.

I would argue you need class C gateways, because
1) 99.9999% of the population can't figure out what a /30 address means
2) Facist countries like North Korea and the USA will need all kinds of
montioring and proxying stuff to "save the children" and "protect software
jobs from piracy" "stop criminals".  All those monitoring devices will 
use up IP space, and I assume they'd have to be installed at each gateway.

And you need at least a class C worth of gateways at each level, because I'm
sure the entire USA has more than 256 under ocean fibers plus microwave
uplink stations, at least.  I suppose New York, NY has more than 250
optical interstate fibers, or will soon enough.

Not hard at all to fill up a 128 bit address space, and we haven't even
tried autoconfiguration based upon MAC addresses and stuff.  You could
literally build a world wide autoconfigured plug and play internet, if you
had maybe 512 bits or so of addressing.

If massively parallel router architecture replaces the current "one big
central router" concept, then all bets are off as to how many ip addresses
would be required.

In other words, in an ideal world, 128 bits still won't cut it, but maybe 
512 bits would be sufficient for a plug and play world wide internet.


Reply to: