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

Re: IPv6 adoption

I landed into the middle of this thread, when it dropped onto the
the IPv6 list, so forgive me if I miss the point (I'm not even sure
which way you're arguing there), but...

Jason Gunthorpe wrote:
> On Sun, 16 Jul 2000, Mika Liljeberg wrote:
> > There are  340 282 366 920 938 463 463 374 607 431 768 211 456 IPv6
> > addresses. That's roughly 313 million addresses per every cubic
> > millimeter of Earth. Granted, the address space is sparse, but I
> > still think it's quite a lot.
> The point here is not that there are alot of addresses, but that routing
> information is stored in the address!

Sure, that's what I meant by the address space being sparse. The
whole point in having aggregatable addresses is that the routing
info is encoded in the address. A router only needs to look at a
small string of bits in the address in order to make the routing
decision. This keeps the routing tables small and makes the routing
decision very fast. This is IETF honoring Moore's law and making
sure that the Internet routing infrastructure doesn't collapse
under future traffic. Compared to that, the risk of running out
of IPv6 addresses is quite remote.

Anyway, having aggregatable addresses is simply a matter of
assigning addresses so that they are tied to network topology.
The routing information doesn't actually eat up any "extra bits"
in the addresses.

> For instance, if you have a router that goes from a leased line to
> ethernet then it no longer needs ARP at all, it just rips the lower 64
> bits off the incoming packet it and slaps it onto the ethernet header
> [with a bit of mangling to go from EUI-64 to MAC].

Can't do that. There's no guarantee that the interface id has been
constructed from an Ethernet address and even if it has, the
destination might be proxied. The router has to do ND as usual.

> Things are similar at all layers of the network, straight through to the
> frame relay, ATM and SONET layers, AFAIK.
> Another simple example is one of the IPv6 in v4 auto tunnel proposals. The
> idea is to encode the IPv4 address of the border gateway in the sites
> IPv6 prefix, this way any other site who wanted to talk to this one could
> do so without doing a IPv6 routing lookup - automatic tunnels.

Err... I have no idea why you mention this. The idea behind automatic
tunneling is simply that you can connect isolated IPv6 subnets together
through an IPv4 network without having to configure tunnels manually.

And yeah, transition mechanisms eat up a minuscule part of the IPv6
address space. That is just to be expected. It's not a big deal.

> I've heard it is yet to be shown how well the other 61 bits of global
> routing information will work, the general idea was to encode broad
> routing information so that backbone routing tables could be made much
> smaller..

Why wouldn't it work? It couldn't possibly be worse than IPv4. Five
ago, when CIDR was introduced in a hurry, Internet backbone routers were
on the brink of collapse. Now they are applying the same principle in
IPv6 and suddenly it's not a good idea?

> Jason



Reply to: