v6 subnet size for DSL & leased line customers
jmaimon at ttec.com
Mon Dec 24 17:10:28 UTC 2007
Scott Weeks wrote:
> Disclaimer: I'm still very much an IPv6 wussie... :-)
> But even in 2000 the policy was and still is:
> /128 for really a single device
> /64 if you know for sure that only one single subnet will
> ever be allocated.
> /48 for every other case (smart bet, should be used per default)
> I work on a network with 100K+ DSL folks and 200+ leased line customers, plus some other stuff. The leased line customers are increasing dramatically. I should plan for a /64 for every DSL customer and a /48 for every leased line customer I expect over the next 5-7 years?
Same disclaimer as above. But perhaps thats a benefit, allowing the
landscape forest view instead of the tree one.
Seems like everything good and desirable in ipv6 was backported to ipv4,
including router advertisements (which nobody uses, since DHCP [yes dhcp
can/could be made redundant] is far far preferred, even by SOHO vendors).
All except the 4 x bitspace.
If it hasnt been backported after all this time, its likely either
undoable or unusable.
Since its quite likely that a minimum 50 year lifetime for ipv4 looks to
be in the cards, judging by bitspace, ipv6 should be engineered for 200
(or 50 to the 4th which makes 125000).
One would suppose that the way to do this is to do as much as is
neccessary to comfortably move the world onto it AND NO MORE. We are not
prophets. We dont even know how many prefixes the average router will be
able to handle in 10 years (considering that a maxed out pc-as-a-router
can handle millions more than the nice expensive 7600), let alone 50.
So the first thing we do is:
Make it as big for ISP's as ipv4 was for end users, by assigning /32
prefixes, minus all the special purpose carvings.
To make things simple, a 4 byte AS should come with a /32.
Brilliant. We have forward ported ipv4 scalability onto ipv6.
For what? So that end users can have nanotech networks? It goes without
saying that I will want my nanotech network(s) firewalled (and natted
for good measure).
Autoconfiguration doesnt require 64 bits. We have autoconfig for ipv4,
it appears to only need 16.
As stated, we dont want people to be taking their /64's with them as
they change ISP's, so imbuing all this uniqueness and matching it with
their global id's and telephone numbers is just asking for trouble.
Unless the whole world becomes an ISP. Presto, address shortage unless
massive depopulation occurs over the next couple hundred years.
We should not pretend to be building an allocation structure that will
not simultaneously satisify uniqueness, portability and scalability for
the next hundred years or so when we clearly are not.
Whats the current state with PI in ipv6? How often will it change?
We could have reserved 90% of the first 32 bits, use the next 32 bits to
assign to ISP's /64 bits, and allow the ISP's to assign an internet
worth of customer their own internet.
Tiered routing? Geo-location routing? All easily made available with
another bit or two from the first /32.
Oh and the whole protocol is still useless, since proper connectivity
to the ipv4 network without an ipv4 stack seems to be somewhat non
standard. Obviously, nobody rolling out ipv6 due to address shortage is
going to tolerate that, and interop strategies will be used, standard or
Expect the interop strategy to be the one with the lowest network
resistance. Thats nat.
IPv6 is a textbook second system syndrome. We could have all been on it
already without the dozens of super-freighters attached to the 128bit
More information about the NANOG