On Tue, 05 May 2009 09:13:06 -0400, Joe Greco <jgr...@ns.sol.net> wrote:
No, it's not too late to make simple changes. We're still figuring out
lots of bits about it.
Yes, it is too late. IPv6 as it stands is a huge pile of crap and bloat.
We'd be better off straping the whole mess and starting over, but that
ain't gonna happen. Over the next dozen decades and hundreds of RFCs, we
might have something that looks like it was designed by competent people
instead of glued together mess we have today that was created by
committees with multiple personal and political agendas.
On the other hand, can you *guarantee* that it will not?
Yes. Yes, I can. Ethernet has been around for decades from 10M to 100M
to 1G to 10G and now we're working on 100G. Look around the room and
count the number of devices containing ethernet technology. It's f'ing
EVERY. WHERE. Every single piece will have to be replaced to support
EUI-64. It's grown into the silicon, so there's no amount of software
updates that can fix it like we're attempting to do with IPv4-to-IPv6.
And quite frankly, you've already conceded that
a technology - firewire - exists that does use EUI-64.
True. But you ignore the fact that firewire isn't used as an internet
transport technology. Where's the 24 or 48 port firewire switches? You
can run IP over fibre channel, but I don't know of anyone who does so
outside of private (read: internal) networks. (Clusters often use FC-IP
within the SAN for node-to-node signaling.)
Ethernet won. It uses 48bit addressing. It's not going to change. That
"mistake" is now cast in diamond. The world is not going to throw away
all the ethernet gear because someone wants to change the addressing
scheme.
Do you have an equally brilliant but completely different suggestion as
to how to implement reliable stateless autoconfig in IPv6?
Sure I do. And I'm not the only one. In fact, many IPv4 systems have an
address generator... the thing that builds "local" 169 addresses.
The simple fact is they took the dirty, brainless simple path of using
what is supposed to be a unique identifier (Layer-2's MAC-48) and directly
attching it to the layer-3 (IPv6) address. Everyone is confusing
"stateless" with "constant" and "consistent". SLAAC doesn't need to
generate the exact same address everytime the system is started.
Stateless simply implies a host is not depending on data maintained from
an external source. A host can generate whatever random number it needs.
It doesn't have to be *globally* unique; it only need be *locally*
unique. There are plenty of ways to generate and verify local uniqueness.
No. Do we have to do that before we figure out what to do next?
Do we have to replace trillions of dollars in hardware because of a
problem we don't have?
Are we too stupid to learn from the period of history we're going
through right now? With IPv4, we've waited until we're just about
out in order to figure out where to go from here. That was dumb.
Predictable but dumb. Why wait for resource depletion in another
realm, when we already know that's a bad thing to do?
You must be new here. IPv6 was designed a long time ago. Long before we
"ran out of addresses". Nobody has deployed it because nobody has
deployed it. IPv4 works. We still have address space to hand out -- and
will for several more years. IPv4 will *continue* to work long after IANA
has no more blocks to assign.
Bottom line... there's no pressing reason to make the jump, and a whole
bunch of reasons to hold off. But you don't seem to care about any of
that -- we should all continue driving our pintos with the exploding gas
tank until your local shop has time to replace it. No. Thanks.
RFC3041.
Ah, so you conceed there *are* ways to generate addresses that aren't the
MAC address. Therefore, they don't have to be 64bits. However, it's
easier to be unique with larger numbers.
You don't think that the IPv6 designers thought long and hard on that
very question? You're second-guessing them? I'm sure we'd all
appreciate a
presentation as to why 128 bits isn't enough.
I'm not guessing at all. I know they didn't. And where the f*** have I
ever said 128bit isn't enough. My whole issue is with forcing people into
0% utilization of their address space "because we have lots of address
space" and "eventually we'll need that space." Yet, you seem to think
we're justified in giving people billions upon billions upon billions of
addresses because we might, someday, somehow, have millions of gadgets
that need to be globally addressable. But that's completely different
from the mess we have with IPv4... handing out /8's because we could, then
throwing on the breaks and promoting (even demanding) "responsible use",
all the way to today where everyone asks for more address space than they
currently need because "we might need it later" but later never comes.
128bit addressing is uber-plenty and will last us a long time as long as
we continue to practice "responsible use".
These are huge numbers that we're talking about. At the time IPv4 was
created, people were looking at 4 billion and refrigerator-sized routers
and thinking, "this'll last us for a while." And it did. But I don't
recall them assuming that IPv4 was the end of the road.
And you don't see the repeat with IPv6? *sigh* I see it everywhere...
the address space is *HUGE*. there's no way we'll ever use it all.
"enough addresses to assign every grain of sand on the planet it's very
own..." But yet, day one we slice the address space in half and place a
"globally unique" (probablly) number in the lower half. And then propose
slicing the uper half into chunks large enough to give every house 256 to
65,536 *individual* globally unique spaces.
You're not being locked into it. Nobody's forcing you to use it. I'm
sure that you can find someone willing to delegate you a single /64 for
you to subnet along the lines of the traditional IPv4 ways.
Yes, we all are. We will all be given a minimum of a /64, while no one
has a need for even a billionth of that space, and aren't likely to for
the forseeable future. When they do, *then* give them the space they
need. Ah, but "renumbering is a pain", you say. That's another of those
IPv6 fundamentals... renumbering your network is supposed to be easy --
prefix delegation and autoconfig makes it all Magic(tm).
