IP addresses through 2024
potaroo.net
potaroo.net
A single static ipv4 address is a /32 slice into the ipv4 space and is considered a reasonable to size give out to a single person or even a small business that asks for it. Of course larger companies and telecom operators need larger network allocations and they have gotten them for many years in the past.
Now realize that if a /64 ipv6 network is the minimum size like a single ipv4 address then you see that the ipv6 address space has 2^32 /32 ipv6 networks. Now with ipv6 any technical person and acquire a entire ipv4 internet sized network in a continuous (globally routeable if you want) range.
And if any sized business today can expect buy a single "class c" /24 range of ipv4 it makes sense that large global compaines get a ipv6 /20 network to run their entire network on it.For example cloudflare[1] uses 6 regional /32 networks and a /29 network for all their routing needs. Imaging trying to build cloudflare with less than 32 addresses in a single /24 ipv4 allocation.
ipv6 is so large that you can just design your network without worrying about subnet size and route based on real policy or security boundaries alone. We will run out of MAC addresses before we run out of ipv6
There are enough IPv6 Addresses for 4.77 x 10^28 for every living person.
If each IPv6 address was a grain of sand…
That’s 2.39 × 10^18 of addresses per person, or roughly enough sand sized addresses to equal about 1.8 times the volume of earths ocean per person.
At sand scale all IPv6 addresses would take the volume of 12 sols.
Conversely, all the IPv4 addresses in this sand scale would slightly over fill an oil drum.
From “IPv6 Addresses: Big Numbers, Big Solutions”: https://www.osti.gov/servlets/purl/1365362
Allocating all addresses to Earth seems very shortsighted.
(Some sarcasm should be assumed.)
It doesn't really matter how you explain large-number math to people who are bad at large-number math (aka all of us) - there's always some bright-spark who misses the point.
The first martians are likely to run their own local network and then use a VPN over DTN for their everyday communications by email (and appear coming from a single IPv6 address).
One good thing though: since most of the humanity’s knowledge is going to be packed in an LLM they won’t really actually need internet to learn about things. But lack of videos may be annoying.
We can give every person a million local networks, with unlimited devices on each local network. That's more accurate and also easy to think about.
Similarly, it’s better to think of the depth of a network’s topology. The size of the tree when full is immaterial, especially when the last 64 bits is intended to be so sparse that random address assignment is viable.
So instead of talking about there being 2^128 addresses, it’s more useful to talk about there being 4k global regions (/12 RIRs), assigning /32s to ISPs in their region.
Each ISP can assign a /48 to each customer site, which can then subdivide into 256 buildings with 256 VLANs each (or some other balance of the these, eg 8 buildings with 64 floors each and 128 VLANs on each floor) with the hosts selecting randomly (or SLAACly) from the final 64 bits.
Some ISPs where customers are in a single dwelling will only give you a /56.
But the allocations for a single device are so large that it's not actually as large as it seems to be. Kinda like usb N only working at N-2 speeds...
It's really hard to comprehend how many unusable ipv6 addresses are.
Having more than 2^16 hosts on a subnet is pretty much impossible. Sticking with "grain of sand" units, but using volumes all from wolfram alpha:
There's 2^80 usable IPs in the entire ipvv6 space, because of the /64 subnets. That's plenty. But for every subnet, that's 40 cubic metres unusable for every subnet, and a sphere 1mm wide of usable addresses.
My ISP give me a /48. I have under 30 addressable devices over 3 vlans. I'm using 40 cubic miles of space. A ratio of about 10 trillion:1
But that's nothing. The IP allocators are happy to give a bank a /16, or in your "grain of sand" measurements 30 times the volume of the moon.
To match my unused:used ratio of 10 trillion to one, chase manhattan would need 2^70 devices, which is billions of addresses per cell.
All the space that fanboys go on about is almost all unusable, so the extreme examples don't really help at all.
We will never run out of IPv6, but I believe in my lifetime you won't be able to get new v6 allocations anymore because they will have all been handed out to the large corporations with deep pockets. Oops, no addresses left - but you can rent one from a cloud provider for a few dollars a month.
IETF draft standard of IPv8: https://datatracker.ietf.org/doc/html/draft-pouwelse-trustch...
Bonus: "De-DSI" == LLM- based fuzzy search + IPv8
So besides the bad design ipv6 is also badly managed?
Capital One would never get a /16 in RIPE.
The IPv6 designers already thought of that and prepared for that eventuality. Only 1/8 of all available IPv6 addresses (the ones starting with the bits 001) is available for allocation under the current rules. If and when that gets exhausted, there will still be more than 5/8 of all IPv6 addresses available to be allocated under more strict rules. (See https://www.iana.org/assignments/ipv6-address-space/ipv6-add... and https://www.rfc-editor.org/rfc/rfc3513.html#section-4 for more details.)
[0] - https://blog.benjojo.co.uk/post/class-e-addresses-in-the-rea...
Yes yes yes, all sorts of really super serious policies were put in place to prevent abuse of IPv4 allocation rules too.
I have a /32 as an individual that I also got no questions asked.
It’s more IPv6 space than I could ever use.
However, IPv6 space has to be measured in /48s, and IPv4 in /24s, as those are the smallest possible BGP announcable blocks.
With IPv6 there's also an unwritten rule that RIRs will only assign blocks ending on a nibble boundary (with the exception of RIPE and their /29s that they expand by one bit every time you run out).
So a /44 is on the second smallest nibble boundary, making it the second smallest allocation possible.
Well, if the VPSes are in seperate sites, at least one per VPS, which means at least 2, which means that according to allocation policies, at least a /44. :)
SLAAC works on any device that supports DHCPv6, though, so you might as well use it if you don't have any requirements to favour DHCPv6
If you rent a vps with a ipv6 “address” you will see that you are given a /64 network.
There is no effort to “fix” the bottom 64 bits because it’s not needed. There are enough /56 networks to for the next 500 years and 99% of people can do everything they need with an isp provided /56. IPv4 has survived with 48 bit nat for this long so 64 bit is even more buffer space
Unfortunately, I've seen multiple VPS providers that will assign you a single address within a /120 or something ridiculous like that.
I presume whoever runs the network side of things with those VPS providers has never used IPv6 themselves or they must be _extremely_ stingy.
Now, of course, a /28 in IPv4 is a tiny range, while a /28 in IPv6 is huge. And there are, of course, a lot of /28s. The fear is only that we may start seeing larger and larger allocations for no good reason whatsoever. Apparently already some random private company (Capital One) has received a /16 IPv6 range - this is an absurd allocation that should simply not be allowed if we don't want to have to move to a new IP version in a few decades.
It is not the limitation, but router hardware is.
Now problem mostly eased, thanks to Moore law, but 20 years ago router memory need to store prefixes-routes graph was very expensive and have some problems with grow.
For example, I know people, bought very expensive server PC boards (but cheap compared to enterprise networking hardware), with DDR2 capability, because it was fastest at that time and promises large RAM space.
Imagine their frustration, when they seen, this hardware cannot allocate prefixes list for their installation (just not enough RAM) and they have to buy new hardware with DDR3.
This even caused "curse of /24", when some providers decided to just discard /24 prefixes and work only with /16 directly and for smaller allocations use some default route.
And this is just because nature of Internet, which is by definition "network of networks", mean, in ideal case all networks should have peer connections at least to all first circle neighbor networks, but better if have peers to few first circles, plus connections to large local hubs.
And now problem become harder, because of grow of p2p solutions, which by definition avoid hub model and use all peers directly.
BTW, colleague from telecom share latest problem - few days ago appear huge traffic. In very short time found source - after appear new "Stalker" game, their developers made huge number of fixes, so fixes are larger than installation, and they somewhere bought so powerful distribution network, distribution of fixes flooded all connections for few days.
One way of thinking about it:
* Stars in the Milky Way: 400 Billion
* Galaxes in the universe: 2 Trillion
So (4x10^11)x(2x10^12)=8x10^23 stars in the universe.
* Size of IPv6 address space: 3.4x10^38
Find the ratio between addresses and stars:
* 3.4x10^38 / 8x10^23
IPv6 offers about 430 trillion times more addresses than estimated stars in the universe. From Tom Coffee's presentation "An Enterprise IPv6 Address Planning Case-Study"
* https://www.youtube.com/watch?v=7Tnh4upTOC4
Perhaps in more human terms: on the surface of the Earth (including oceans), there are 8.4 IPv4 addresses per km^2. Not counting the oceans, that would be 28 IPv4 addresses per km^2 land.
IPv6 gives 10^17 addresses per mm^2 (yes, square millimeter).
In terms of volume, 10^8 IPv6 addresses per mm^3 throughout the Earth.
Another way of looking at it:
*math property: x^y = x^(a+b) = (x^a)x(x^b)
* IPv4 addresses are 32 bits (2^32)
* 2^32 ~ 4.3 billion
* So the IPv4 Internet has ~4.3B devices on it
* IPv6 subnets are 64 bits, /64 (2^64)
So, an IPv6 2^64 subnet is the same as (2^32)x(2^32), which means (4.3B)x(IPv4 Internet). I.e., a single IPv6 subnet can hold the equivalent of four billion (IPv4) Internets.
For general interest, this is (among other accolades) 2012 Internet Hall of Fame Inductee Geoff Huston https://www.internethalloffame.org/official-biography-geoff-...
(Legally they're supposed to use those addresses within a year(?) of buying them, but I won't pretend that anyone would really notice whether that's the case or not.)
I've got a /24 that I obtained back in the Wild West days of the early 1990's, when all you had to do was send an email form and a few minutes later you'd get one assigned. I haven't used it in over 20 years now, but nobody has ever come to take it from me.
Ofc, server like desktop applications should randomly choose an IPv6 there, but it has to stay stable since this number will have to be given to other people to connect to such server (I am talking dodging DNS $$$ racketeering or Big Tech "name<->IPv6" mapper service, aka for the smhol internet).
Only if the ISP is providing a /64 prefix ofc.
In my country, nearly 100% of domestic internet lines are working like that, and it has been the case for years. The main issue, is IPv6 on mobile internet: in my country most mobile internet has IPv6 enabled... but it seems you don't get a /64 prefix but a different /128 ip address at each authentication of your sim card.
If it ends up not being a trick of my current IPv6 mobile internet modem, this is very bad, REALLY bad: I cannot give a stable mobile internet IPv6 to my contacts for communication (sort of a phone number dedicated to them). Mechanically, it will force people to use classic centralized client-server services, hence force people into Big Tech.
> The use of NATs force the interactions into client-initiated transactions... (abbreviated)
There is absolutely ZERO chance that that, let's say Roomba, is going to let you connect directly to your vacuum robot at your home, from your cell, over ipv6, without going through their proxy server in an AWS Datacenter in Virgina.
The nativity of engineers is face-palm inducing.
Roomba will _never_ give up that control over your device. Same with your Tesla, your iPhone, your security cameras, or your ring doorbell. Zilch, none, nada. Giving you direct control, even if ipv6 were fully implemented, is simply not on the roadmap for the companies. They want to control you. They want to control your devices. They will not release the death grip on this, as it releases their control over your property post purchase.
This sounds incredibly cynical but it's playing out in front of us. By restricting the use of their services, they now control and downstream resale of the device and can force people to buy new, rather than repair or renovate. And of course the leaders in all of this anti-repair initiatives are the so called "Green" companies of the world.
You are doing a "there can only be one binding constraint" analysis, and I don't think that is get. Yes, a lot of incumbents want client-server only, but IPv4 still gives them power enforcing the status quo. If we had IPv6 everywhere, they would need to try harder to enforce it, and we would have to try less hard to change it.
If I recall correctly, IPv6 routing was supposed to be hierarchical based on ISP and region, and that would help routing tables. But what if companies in the US, for example, buy a /19 and then divide it up and use it across the globe?
I assume routers will have to come with a lot more RAM.
The Internet gave up on that a while ago. IPv6 routing works like IPv4 where people just advertise whatever routes they want. Router RAM has been fixed with newer HBM-based ASICs supporting millions of routes.
A combination of “getting there first” (and thus, a bunch of decaying infra that doesnt support v6- and needing it all updated) and the “free-net” thing where you can choose your ISP in some apartments.
For those not in the know: internet is usually negotiated by your entire apartment building, but some (especially the largest rental building association “MKB”) give you the choice to choose your provider.
The way it actually works is complicated, but nobody will invest in this “open networks” system, so it is stuck in time fully, and even providers (like Bahnhof) who want to give v6: can’t with that system.
It seems hard to find info on the open network stuff, I know it all from a friend working at Bahnhof, here’s at least part of it:
- https://ieeexplore.ieee.org/document/5549139
- https://www.stadsnatsportalen.se/pages/onapi (Swedish)