back to article IPv4 is OVER. Really. So quit relying on it in new protocols, sheesh

Well, that took a while. Eighteen years after the IETF brought us IPv6 as an answer to then-looming-now-upon-us IPv4 address exhaustion, the Internet Architecture Board says: no more. Getting IPv6 into the field has been a long, slow slog. According to Google, IPv6 carries just 14.6 per cent of search requests to the web giant …

Silver badge

Re: Exhaustion? and yet...

Um yes, but if you need a /16? (The equivalent in IPv6 is a /48 which is trivial to get.)

0
0
Silver badge
Pirate

Re: Exhaustion? and yet...

"Try getting a /24 it is pretty painful."

Unless you're a college/university. I got one in 2013 and it was way easier than the tech guru at my ISP had warned me about. ARIN didn't really even ask for justification, they just saw I worked for a university and said "Here you go..". Sometimes it's nice to get an unexpected break...

1
0
Silver badge

What new protocols are IPv4 ONLY?

I can't even think of how that would happen, unless you define in the protocol that an address must be 32 bits, or four octets in text separated by periods. Anyone have an example of a new protocol crafted in the last five years with such a restriction?

3
0
Silver badge

Re: What new protocols are IPv4 ONLY?

Reading the IAB statement and RFC 6540, it would seem the big problems are: Firstly, the need to review, mainly the pre-IPv6 RFC's, to ensure they correctly refer to 'IPv4' rather than 'IP' - given when they were written 'IPv4' was 'IP' everything else was 'experimental'. Secondly, it seems the master RFC that defines the current suite of RFC's that define the current IPv4 & IPv6 profiles and thus the current "Internet Standard" RFC's needs to be updated to take account of IPv6...

As for the request: "The IAB expects that the IETF will stop requiring IPv4 compatibility in new or extended protocols.". In most cases, this won't be an issue, however, for as long as IPv4 is widely used I anticipate there will be the occasional need for an IPv4 specific RFC to address some issue, such as security.

Otherwise, the only two areas, within the IETF remit, where new protocols might be written as IPv4 only are:

1. Those intended to enhance the IPv4 network layer protocol suite, namely protocols designed to carry IPv4 addresses etc. as part of their payload.

2. Link Layer protocols that assume they only carry packets and hence need to interface to an IPv4 network layer.

Naturally none of this prevents protocol implementors and application programmers from using the various API's in ways that only support IPv4 or IPv6.

0
0
Bronze badge

Not exactly important

IPv6 does solve the address space problem but it does it in such a cack-handed way that its no surprise that its been a bit slow to be adopted. Since they're both just containers for higher level traffic there's no particular reason to favor one or the other, especially if you're running all your traffic in a private tunnel like a VPN.

4
0
Silver badge

Re: Not exactly important

"but it does it in such a cack-handed way"

The original proposal for IPv4 was to use 128 bits for addresses. It was cut down to 32 bits because IPv4 was intended to be a kludge with an expected lifespan of a decade at most.

0
0
Anonymous Coward

IPv4 and IPv6

Speaking from experience, the problem I see with the latter is that the standard is perceived to be massively more complicated than IPv4's for little additional practical benefit. I imagine that if they had instead just increased the address space without changing anything else (conceptually speaking), adoption would have been much faster and IPv4 would be just a footnote in history books by now.

8
0
Silver badge

Re: IPv4 and IPv6

> I imagine that if they had instead just increased the address space without changing anything else (conceptually speaking), adoption would have been much faster and IPv4 would be just a footnote in history books by now.

Probably true, but then we'd also still be having to work around some of the issues that are "solved" in IPv6 - as in, they looked at what the problems were and didn't design in the same problem again !

Take one "simple" example - determining if another node is "locally connected". If your only experience is with small networks, ethernet, one subnet per network domain - then you'll be thinking that this isn't a problem (and I was in this camp too until the penny dropped). Simple solution, you look at your IP and the other node's IP, and if they are in the same subnet then you are "locally connected" (meaning that you can send a packet directly to the other node).

Trouble is, this isn't true in many networks. Take a wireless mesh for example, and I believe cellular systems can be similar - it needs a flat IP space so devices can move around freely, but propagating all that ARP broadcast traffic would be very wasteful, and so the nodes have to effectively "fake" the ARP process or bu**er about with routing tables to make what the end device sees as a flat network into something that's actually routed. One example of "IP in same subnet, isn't locally connected".

And then there are cases where you've added a second IP subnet to a network - a shared network, often done when you run out of addresses. Now you have a case of (unless you manually add routing rules to each device) "device in different subnet is actually locally connected".

In IPv6, the routing information provided to each node by the network has not just a list (note, list which may be one or more than one) of prefixes on the network, but also information on which of these are to be considered "locally connected". Thus the process is a bit more complicated than in IPv4, but it inherently supports all four combinations of "same prefix/different prefix" and "locally connected/not locally connected" while IPv4 only supports 2 of them without messing about.

This is one example where IPv6 appears more complicated (it did to me before I got my head around it), but which is there for a valid real-world reason.

Most of IPv6 isn't hard - it just looks daunting. Perhaps part of the problem is that there isn't a good online tutorial (at least I haven't found one) that can take someone with "basic IPv4 skills" and take them gently through a controlled learning process. Looking around, I've found lots of stuff, but a lot of it is either too basic and doesn't explain the "WHY ?" or too complex and anyone not already familiar with the subject matter is just going to drown.

The nearest I've got is to do the IPv6 certification at TunnelBroker.net. I think it still leaves a lot out that you need to know (at least if you are a serious network nerdtechie), but it at least has a progression of steps with tests at each step to show you've grasped things so far.

1
0
Silver badge

Re: IPv4 and IPv6

Another reason for IPv6's structure was to deal with routing table problems. Thanks to all the IPv4 address hawking, there's no longer any logical, down-the-bits progression of topology. That's the reason IPv4 routing tables have gotten so huge: to the point some routers choke at the size. The massive overprovisioning of IPv6 is intended to stem a repeat of this for as long as possible, much like how ZFS's filesystem provisioning is intended to make sure physical limits hit before the logical limits do.

1
0
Anonymous Coward

Re: IPv4 and IPv6

> Probably true, but then we'd also still be having to work around some of the issues that are "solved" in IPv6

Yes, fair enough, but at least those were familiar problems which already had a solution or at least a workaround. I think I would have opted for a more gradual evolution, given the chance.

> Perhaps part of the problem is that there isn't a good online tutorial (at least I haven't found one) that can take someone with "basic IPv4 skills" and take them gently through a controlled learning process.

Nearly eighteen years on since RFC2460 was published, makes you wonder why indeed, doesn't it?

2
0
Anonymous Coward

IPv6 is OVER!

ISO foisted a dumb, over-designed network protocol on the world, called OSI. Laughingly, the 'L' in LDAP stands for lightweight. The world overwhelmingly adopted the five-layer model of IP, which was succinct, well-architected and designed to interoperate. OSI died a death.

Apparently, the OSI people got the last laugh and delivered IPv6. 18 years later, the entire planet has collectively said "uh, no thanks..." To put it into context, if IPv6 were a mobile phone technology, our phones would have four times as many buttons for dialling numbers, phone numbers would be 40 digits long, the phones themselves would look like Motorola units from the 80's, and the kicker? They wouldn't work with *any* of the existing phones. "Yes, you too can be the first kid on the block to have an incompatible phone. You can wait until all your friends have converted, before you can actually call them..."

Instead of flogging this dead horse which has failed to get any significant market traction in almost two decades, they should admit defeat and develop IPv8, something with a maximum of 64 bit addressing, and something which seamlessly interoperates with IPv4. In reality, you don't need to change the upper layers of the stack; leave TCP and UDP alone. Just up the bit count for addressing, bearing in mind that every additional bit of address doubles the number of addresses.

4
6
Silver badge

Re: IPv6 is OVER!

And keeping in mind adding just ONE bit breaks IPv4 compatibility. Completely. Period. Because the IPv4 stacks expects 32-bit addresses. Not 31 or less, not 33 or more, 32. It's hard-coded into the spec and can't be changed or extended.

IOW, the ONLY way to add more addresses is to break the mold. If we're going to break the mold, why not start from scratch with some future-proofing in the process. We're using 128 bits (1) so we can keep logical down-the-bits routing that keeps routing tables small and keeps routers from choking, and (2) to help keep the IPv4 exhaustion problem from happening again down the line as unexpected things (like, oh, more and more people starting to connect EVERYTHING to the Internet like what's happening RIGHT NOW) take up more IP space. It's the same reason ZFS uses 128-bit provisioning even if it could probably get away with 96 or maybe even 64 bits.

0
0
Anonymous Coward

Re: IPv6 is OVER!

...The world overwhelmingly adopted the five-layer model of IP, which was succinct, well-architected and designed to interoperate. OSI died a death...

Apparently, you have absolutely no idea of what you're talking about.

2
0
Silver badge

Re: IPv6 is OVER!

"phone numbers would be 40 digits long"

When phone numbers were standardised so that 11 digits was the maximum needed to dial almost all international numbers, it was regarded as excessive.

13, 14 and 15 digit numbers are popping up all over now and the world's phone routing network is so messed up it makes the BGP4 tables look tidy.

0
0
Anonymous Coward

Re: IPv6 is OVER!

> To put it into context, if IPv6 were a mobile phone technology, our phones would have four times as many buttons for dialling numbers

One of my professors used to say that if you have to use an analogy to explain something, you do not understand the subject well enough.

0
3
Anonymous Coward

Re: IPv6 is OVER!

> It's the same reason ZFS uses 128-bit provisioning even if it could probably get away with 96 or maybe even 64 bits.

If it were up to me, I would design a 73-trit architecture. Purely because a) I like primes, and b) would give me an excuse to use Łukasiewicz logic.

Off to kickstarter, back in a bit.

1
0
Anonymous Coward

Re: IPv6 is OVER!

And keeping in mind adding just ONE bit breaks IPv4 compatibility. Completely. Period. Because the IPv4 stacks expects 32-bit addresses. Not 31 or less, not 33 or more, 32. It's hard-coded into the spec and can't be changed or extended.

So add an extended header at the start of the data packet, put the extra address bits somewhere in that, and define some special SRC/DST values for the existing fields that say 'this is an extended v4 packet'. Existing v4 stacks will route based on the existing address fields so that "real" v4 packets buzz about as normal and "extended" v4 packets get routed transparently to some new dual-stack systems that can handle the extended header. Systems that can fully process the extended headers can use them to reach new systems that don't have old v4 addresses.

That way dual-stack systems can talk to anyone, v4 systems can still talk to v4, and v4 networks can transparently route between islands of "extended v4".

OK, the extended header idea needs to play tricks, so its a hack that offends the purists, but so what?

2
1
Anonymous Coward

Re: IPv6 is OVER!

13, 14 and 15 digit numbers are popping up all over now and the world's phone routing network is so messed up it makes the BGP4 tables look tidy.

But no-one has had to buy a new phone to use it, the complications are handled transparently by the networks.

1
0
Anonymous Coward

Re: IPv6 is OVER!

Maybe you'd like to explain what Layers 5 and 6 of the OSI model map to in an IP world. Or perhaps, you don't actually know what you're talking about?

0
0
Silver badge

Re: IPv6 is OVER!

"So add an extended header at the start of the data packet, put the extra address bits somewhere in that, and define some special SRC/DST values for the existing fields that say 'this is an extended v4 packet'."

Which does NOTHING for the "old device talking to a new device" problem, which IPv6 also has. No IPV4-only device can talk to anything outside the scope of IPv4 without a proxy. It doesn't matter whether it's IPv6, your idea, IPX, or whatever. IPv4-only devices can ONLY see IPv4 devices, and this cannot be solved in the scope of IPv4 alone. So again, if you have to break, why not break clean?

1
2
Silver badge

Re: IPv6 is OVER!

"One of my professors used to say that if you have to use an analogy to explain something, you do not understand the subject well enough."

Yes, all that endless row of people who ended up having to try to explain voltage and current using liquid flow analogies to an audience who's eyes immediately glazed over on any attempt to explain the subject directly must all have been ignorant on the matter themselves. Definitely. Yeah, that must be it.

1
0
Silver badge

Re: IPv6 is OVER!

"Yes, all that endless row of people who ended up having to try to explain voltage and current using liquid flow analogies to an audience who's eyes immediately glazed over on any attempt to explain the subject directly must all have been ignorant on the matter themselves. Definitely. Yeah, that must be it."

IOW, YOU try explaining electricity to people who can't understand why 10mA (or whatever current you can get from a handheld stunner) can shock them off their feet.

1
0
Silver badge

Re: IPv6 is OVER!

"But no-one has had to buy a new phone to use it, the complications are handled transparently by the networks."

Because telephones are pretty dumb devices. All the smarts are done by the exchange. However, IPv4 devices can't do that. There's a certain minimum degree of complexity involved. And one of the catches is that IPv4 devices expect a 32-bit address, period.

0
1
Anonymous Coward

Re: IPv6 is OVER!

> Yes, all that endless row of people who ended up having to try to explain voltage and current using liquid flow analogies

Yes. And Einstein can talk about trains in the context of relativity all he wants, but we're not all Einstein (and his wasn't an analogy anyway).

0
0
Bronze badge
Happy

Leftover Halloween Decorations

Very ingenious to use those leftover Halloween decorations before you pack them away for another year.

Is there anything else dead you'd like to make us aware off, besides the brains of the political leadership? It's the perfect time.

0
0

Granny factor

I won't be the first to opine this, but the worst thing about v6 is the lingo.

I (eventually) got the heads of my dependents, both at work and closer to home, around the concept of an IPv4 address. Four integers with dots in between. Sort of like a phone number, in that the format was consistent.

IPv6? 8 numbers in hex notation (with which we're all super comfortable, yeah), colon-delimited, which if they happen to be zero can be omitted, and if a number of them adjacent to each other are zero the colons can also be omitted just to make it extra opaque.

Do people really not get this? It's too much. Joe sixpack does not want to speak hexadecimal, nor to make sense of which address components are elided or why. If the address is reliably eight integers, with colons or periods or fucking Euro symbols delimiting them, that he can grok (possibly with help). IPv6 address notation is design by engineers, for engineers; granny need not apply.

5
0
Silver badge

Re: Granny factor

Because it's the engineers who are most likely to actually be fiddling with these numbers. Thanks to things like DNS and DHCP, why would Granny need to even know these numbers exist? If something goes wrong with their Internet, Granny's likely to call the grandson or some tech who would grok this stuff anyway.

In other words, for most people, IP numbers is Somebody Else's Problem.

1
2
Anonymous Coward

Re: Granny factor

"In other words, for most people, IP numbers is Somebody Else's Problem."

Are you sure you're on the correct website? You do realize quite a large part of El Reg's readership are the folks to whom you've just casually tried to give a whole new mess of IP number problems as if it's No Big Deal. We're the "Somebody Else" you're talking about. We aren't all web-designers here, some of us do real work.

1
0
Silver badge

Re: Granny factor

"We're the "Somebody Else" you're talking about. We aren't all web-designers here, some of us do real work."

And we're usually the people who know how to get around things like octets. WE WRITE THEM DOWN. I've been dealing with IPv6 for years now, and it's really been no big deal for me because I'm literate enough to see the differences. For those who don't grok it, they shouldn't have to. For those that do, it comes with the territory.

0
0
Boffin

Yes please

Now, if ISPs can get off their asses and start providing IPv6 addresses, it would be very nice! 18 years and counting ... the protocol is old enough to vote and drink in most countries by now!!!

0
0

IPv15

I sure hope we get it right by IPv15. It's a shame to run out of bits.

(For what it's worth, I was there for IPng and the arguments over variable length addresses to appease the OSI weirdos. This has been a much longer slog than anyone ever imagined.)

0
0

I hate to say it, but it's IPv6 that's over. When does IPv6 have any value? When you can communicate with the rest of the internet without an IPv4 address. Dual-stack HAS NO VALUE. It doesn't save IPv4 addresses. This whole exercise is feel good for lovers of IPV6 who won't accept that they've lost. What's needed? A complete redesign with forward and backward compatibility between IPv4 and IPv7. IPv6 was a series of disastrous decisions and it is blocking the progress that would come by starting over.

0
0
Silver badge

"When you can communicate with the rest of the internet without an IPv4 address."

Working on it right now. There's an IPv4 compatibility space within the IPv6 space.

"Dual-stack HAS NO VALUE. It doesn't save IPv4 addresses."

It's not meant to. It's meant to transition IPv4 into IPv6. Sorta like a ladder so you don't have to climb the sheer cliff with your bare hands.

"What's needed? A complete redesign with forward and backward compatibility between IPv4 and IPv7."

Which is a non-starter. You can't make IPv4 forward-compatible with anything BUT IPv4 because of the hard 32-bit address limit. And no extension to IPv4 will work because older devices won't grok it; they STILL won't be able to see the new addresses, meaning they're STILL left out. You're asking us to cram a baker's dozen in an egg carton (128 bits in a system that can only grok 32). It'd be like trying to perceive a tesseract (a FOUR-dimensional object) in only three dimensions: something WILL get lost along the way. Which means as more sites got IPv6 (and by that I mean IPv6-ONLY--ask Asia), IPv4 devices won't be able to see them without a proxy, which has its own issues. And trying for a sharp break is like trying to go cold turkey on a hard drug: too much risk of withdrawal complications.

As the article says, get over it. The ONLY way to get more than 4 billion devices on the Internet is with something OTHER than IPv4, and since IPv4 can ONLY understand IPv4 (AND its 4 billion device limit), that means we either do it the way we are now or we break the Internet trying something else.

0
0

Page:

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Forums

Biting the hand that feeds IT © 1998–2017