[arin-ppml] IPv6 Heretic thoughts

Iljitsch van Beijnum iljitsch at muada.com
Fri Sep 5 10:59:45 EDT 2008


On 4 sep 2008, at 19:40, Cliff Bedore wrote:

> Having been reading this group for approaching a year or so now, I  
> think
> I've seen the problem with adopting IPv6.  Nobody really wants it.

I don't think that's true.

Most people don't know enough to either want it or not want it. Of the  
people who know, I'm pretty sure a decent number want it, but the  
trouble is that if you're the only one running IPv6, it doesn't do  
much for you. Other people have to run it, too.

As someone how used to configure routers for a living, I want IPv6,  
because it makes my former job easier. (But few people configure  
routers, let alone for a living.)

> The problem:  There is no compatibility bit in IPv6 that says I'm just
> like IPv4 but I have 96 more address bits.

Although it doesn't seem like it, the current situation is as good as  
it gets. For the details:

http://www.bgpexpert.com/article.php?article=106

> The backbone for the Internet will have to
> be IPv6, DNS will have to be IPv6

Not a problem. I already get 10 - 20 % IPv6 hits on my DNS server and  
25 - 50 % of the big backbone networks have some kind of IPv6 running  
today.

> and IPv4 will be treated as IPv6 on
> the Internet and translated through the "converter box (CB)".  This
> means that the CB will have to do both translation and DNS lookups for
> the v4 hosts.

I'm not sure what you have in mind. The problem is that at some point  
within the next years, there won't be enough IPv4 addresses to  
continue current practices. So ISPs will have to figure out a way to  
connect new customers without being able to give these customers their  
own address. NAT can solve the immediate problem, but you can't slice  
and dice public IPv4 addresses forever by adding more NATs.

So at some point end-users will gradually have to start running IPv6.  
At that point, translation becomes useful because these people can  
then talk to people who are still IPv4-only. But translation in and of  
itself doesn't solve anything, and it also incorporates NAT so most of  
the downsides of IPv4 NAT are still there.

> Since there are 64 bits per subnet in IPv6, there will
> never be a subnet that can't split off IPv4 addresses through the CB  
> for
> translation.

Not sure what you mean and no idea what CB is. (Citizen's band?)

> That's a short summary of a big problem but I think it's obvious that
> there has been little real adoption of IPv6.

We don't need IPv6 today. So the fact that we don't see IPv6 today  
doesn't mean all that much. There's still time.

> We really need a program
> that accomplishes what the US HDTV program did.  Tell people that "on
> MM/DD/YY, the Internet backbone will be IPv6 only.  If you want to run
> IPv4, you will need one(or more) of the converter boxes for your IPv4
> addresses.  If you don't do this, you will lose Internet connectivity"

Ah, but our technology is much more advanced than that. Unlike digital  
TV (which may or may not be HD) versus analog TV, we can run both IP  
versions side by side, and even when IPv4 runs out of steam because of  
lack of addresses, the people who already have an address can keep  
using it.

> 1.  All IPv4 space effectively becomes PI space since you can tuck  
> your
> IPv4 into any IPv6 subnet

Unfortunately, this only works in the outgoing direction. In other  
words: you can put hundreds or thousands of clients behind a single  
public IPv4 address and translate from their public IPv6 address or  
private IPv4 address to that public address when they initiate an  
outgoing session, but receiving incoming sessions is much harder: then  
the translator needs to know how to map the sessions and the rest of  
the world needs to know which port number to use (no "80 for web, 25  
for mail" shortcuts).



More information about the ARIN-PPML mailing list