Because there are 'only' 2^32, roughly four billion, IPv4 addresses. There are also over eight billion people on this planet, and many of them have multiple computers (smartphones, tablets, laptops, desktops, etc).

As a 'temporary' measure, Network Address Translation (NAT) was introduced so people could get one public IPv4 address and have multiple machines on the Internet behind it.

The trade-off was the machines behind the NAT mechanism couldn't be reached from the Internet. This isn't a big deal for most consumption-based computing (fetching mail, watching videos), but there are use cases where you want to be able to talk to the other system, so yet another set of technologies had to be invented to allow punching holes through NAT (UPnP, NAT-PMP, PCP, STUN).

But we're at the point in some place where there aren't even enough IPv4 addresses to give to ISP customers even if they're using NAT. So you have a private IP address for your home system, and then your router gets a private IP address on its "public" WAN connection, and you end up going two layers of NAT. There is an entire segment of IPv4 addresses (100.64.0.0/10) reserved for telco use and doubel-NATing:

* https://en.wikipedia.org/wiki/Private_network#Dedicated_spac...

* https://en.wikipedia.org/wiki/IPv4_shared_address_space

We can continue this NAT-upon-NAT(-upon-NAT) silliness, or we can simply make the up-front investment and start pushing out IPv6 in more places. Your router still provides firewall protection, but reachability is a lot more straight-forward because the (potentially double) address translation goes away.

Edit: fix a bunch of typos.

Playing Nintendo Switch online doesn't work if you're double NAT'd unless you go through some router hoops. It seems small, but it's quite aggravating if you're not tech oriented like the majority of the people on this forum.

> NAT-upon-NAT(-upon-NAT)

But there is also a fundamental limit of 2^16 = 65536 concurrent connections, right?

Yes, from one NAT IP using the same protocol to a single public IP using a particular port. Change any of those and you can do 2^16 concurrent connections more per change.

In other words, NAT works on the five-tuple of: protocol, source IP, source port, destination IP and destination port.

> The trade-off was the machines behind the NAT mechanism couldn't be reached from the Internet

That can also be pretty awesome given that so many devices now may not work in the interest of their users.

It is a hassle for IOT, but we have long since then solutions for that and maybe those are better anyway.

My provider only supplies IPv6 and you always connect to IPv4 through a transport. That is sometimes down or has too many concurrent users. Makes the net almost unusable still. Perhaps I should learn Hindi.

You need a firewall if your end points are publicly reachable, not NAT.

NAT still provides some layer of protection, especially if you don’t trust the firewall software.

This is going to open up a can of worms once everybody gets global addresses and can’t figure out how to configure their firewall.

It actually doesn't. NAT just changes the apparent source address of outbound connections; it doesn't do anything to inbound ones.

In any case, people manage to do both NAT and firewalls on v4 today so I don't see why they'd suddenly be unable to do firewalls in v6, especially since you don't have the complication of needing to figure out NAT as well.

The large address space also helps a lot, because it makes it much harder to find servers on v6 (including deliberately exposed servers, e.g. cameras or NASs that people want to access from a different network), compared to v4 where you can enumerate all active servers over the entire internet without much trouble.

NAT prevents an external host from making arbitrary connections to a host behind the NAT without further configuration. This is something your average person doesn’t need to know or really care about. But once every device they have is globally routable they will have to care and ensure their machines are secure and behind a firewall. And if that firewall should fail then you’re sitting on the public network with your pants down.

Your other part is security through obscurity, and I can think of at least 2 ways to scan the entire address space in a short amount of time. So nope doesn’t count either.

It doesn't do that. How can rewriting the source address of outbound connections prevent inbound connections?

You're not going to exhaustively scan the entire v6 space in any short amount of time. It is possible to whittle down the space you need to scan, but only moderately. It's still rather unviable compared to v4.

Pick any network doing NAT, attempt to make a connection. It will be denied. NAT devices are both firewalls and translators and if you don’t have a configuration for port forwarding, be it via established connection or manual configuration, the connection won’t get through. And importantly this is a configuration the user doesn’t touch.

And scanning IP spaces is insanely easy to parallelize and uses so few resources an arduino can be used to scan. Given enough nodes it’s instant. And with every windows box on the planet globally routable, bonets will never be stronger.

But let’s pretend this is true and say it takes too much time. What about when it doesn’t? What happens to your security via obscurity then?

I've tested it before, multiple times; NAT won't deny an inbound connection. A router that's NATing outbound connections will allow inbound connections through unless there's also a firewall. Of course it's very common for there to be a firewall as well, but they're still a separate thing to NAT.

> And scanning IP spaces is insanely easy to parallelize and uses so few resources an arduino can be used to scan. Given enough nodes it’s instant.

You're underestimating how big v6 is. Scanning a single /64 takes ~737 million terabytes of traffic. If you used a trillion Arduinos in parallel you could scan 2^40 /64s simultaneously, and it would only take 1870 years for the scan to complete, assuming that every single one of both the Arduinos and the target networks have a 100 Gbit/s internet connection each. Your power consumption would be... about the same as Italy's, which is actually the most reasonable part of all this (except I assumed the 100 GBit/s network connections would take no power).

> And with every windows box on the planet globally routable, bonets will never be stronger.

You're thinking about botnets that spread by brute force scanning, right? But as mentioned, this will be substantially more difficult on v6, to the point that network scanning won't be a very viable technique for spreading a botnet. On top of that, most Windows machines will be behind two separate firewalls, so I don't see how them being globally routable will make botnets stronger. Given that it'll be harder to find targetable hosts, I'd instead expect botnets to be weaker than ever.

Also, remember that a lot of botnets spread by exploiting servers that were deliberately exposed to the internet. Making these hard to find is the only defence they have, and it's not possible on v4. NAT can't help either, however it works.

An insecure, hard-to-find machine is still an insecure machine, but making it difficult to find vulnerable hosts makes it harder to build a botnet, which leads to a very real increase of security on the internet as a whole. Even if a few machines are found, nothing much happens to the overall security so long as it remains hard enough on average.

Think of it as being something like vaccination for the internet.

> I've tested it before, multiple times; NAT won't deny an inbound connection.

Test it with a consumer level router, stock configuration. I'm not concerned with networks that have split NAT / firewall devices. These are setup by people that know what they're doing.

> A router that's NATing outbound connections will allow inbound connections through unless there's also a firewall.

Yes which is my chief complaint. Not getting rid of NAT, pushing the firewall to the client.

> Scanning a single /64 takes

Why are we the entirety of the 64 subnet? You can use some knowledge about networks to cut this down a lot. Just the one knowing a network is there is enough.

Then you're also assuming people won't group together on this. Lists will be sold in short order, making the search space even smaller.

You're still hiding security behind it being hard to scan because of a large number, so that will quickly be invalidated.

And then if someone say hides in the middle of their /64 and changes their address every hour the attack will then change and it will become focused with malware or other means, except there won't be a fronting NAT protecting them to stop inbound requests.

> Think of it as being something like vaccination for the internet.

What does this even mean? How does a computer get a vaccination?

Alright. Most consumer routers do both NAT and firewalling, especially in their stock configurations, but I can drag one out and disable the firewall on it to test with.

I hooked the router up to my network, and then hooked my laptop up behind the router. The router's WAN address is 192.168.4.101, and my laptop got 192.168.1.9. If I try to connect outwards from my laptop, the connection appears to come from the router's WAN address:

  18:47:04.854781 IP 192.168.4.101.45598 > 209.216.230.240.80: Flags [S], seq 3415044640, win 29200, options [...], length 0
  18:47:04.988960 IP 209.216.230.240.80 > 192.168.4.101.45598: Flags [S.], seq 2572527210, ack 3415044641, win 65535, options [...], length 0
  18:47:04.990322 IP 192.168.4.101.45598 > 209.216.230.240.80: Flags [.], ack 1, win 913, options [...], length 0

That confirms that NAT is working. Next, I'll try to connect to a server on my laptop from outside the router:

  18:49:32.400441 IP 192.168.4.2.58084 > 192.168.1.9.9999: Flags [S], seq 2394060892, win 64240, options [...], length 0
  18:49:32.431718 IP 192.168.1.9.9999 > 192.168.4.2.58084: Flags [S.], seq 3958961179, ack 2394060893, win 28960, options [...], length 0
  18:49:32.432013 IP 192.168.4.2.58084 > 192.168.1.9.9999: Flags [.], ack 1, win 2008, options [...], length 0

You can see it works completely fine. The inbound connection successfully completes even while the router is NATing outbound connections, and this is on a regular consumer router.

> Yes which is my chief complaint. Not getting rid of NAT, pushing the firewall to the client.

v6 doesn't necessarily push the firewall to clients. You can and generally do still firewall on your router.

> Why are we the entirety of the 64 subnet? You can use some knowledge about networks to cut this down a lot. Just the one knowing a network is there is enough.

I know you can cut the search space down somewhat, but it's still massively bigger than v4, and therefore it's still going to be harder to find servers via scanning in v6 than in v4. NAT won't help in the slightest with this.

I'm not trying to suggest that anybody's security should (or even could) rely on hiding their hosts in a big sparse network; you should obviously run a firewall, and pretty much every consumer router does in fact do that. I'm just saying that if somebody does run without a firewall -- or deliberately configures it to permit inbound connections -- the large address space reduces exposure simply by making it harder to find any listening servers.

> What does this even mean? How does a computer get a vaccination?

It means that there's some conceptual similarities between the two situations. With vaccines, it's possible to completely stamp out a disease even if a small percentage of vaccinated people still catch the disease in question.

Similarly, even if a small number of servers are found by brute force scanning, so long as it remains hard enough to do so on average brute force scanning will remain an unviable method of spreading malware. This won't eliminate malware in general (because there are plenty of other infection routes to use) but any malware that relies on exploiting random vulnerable servers is going to have a much harder time in v6, not a much easier time as you assumed above.

> Alright. Most consumer routers do both NAT and firewalling, especially in their stock configurations, but I can drag one out and disable the firewall on it to test with.

Nope firewall on. You're still missing the point. Consumers will disable their client firewall once it interferes with something they're trying to do. A physical NAT device with a firewall won't allow this easy circumvention as those who don't know what they're doing won't generally go mucking around their router settings. They will click a button though, and if they have malware on their machine they can disable it via software means. Firewalls are also software, and just like all software they have bugs. There's an old one where sending certain DNS queries through windows firewall would disable it entirely.

> v6 doesn't necessarily push the firewall to clients. You can and generally do still firewall on your router.

Again you're not thinking of the entirety of the changes here. v6 on its own won't, but the entire point of global link addresses is to avoid NAT and other things. Further, the firewall must remain, but it must be easily and programmatically configured. This leaves pushing it to the client, or leaving exactly what we have today.

> I know you can cut the search space down somewhat, but it's still massively bigger than v4, and therefore it's still going to be harder to find servers via scanning in v6 than in v4. NAT won't help in the slightest with this.

NAT won't, but a firewall will.

> the large address space reduces exposure simply by making it harder to find any listening servers.

Reduces exposure, but does not eliminate it. Agreed though, it is harder to scan ip6 space than ip4, but not impossible

I was under the impression that the point was "NAT blocks connections", and I was attempting to point out that it doesn't. Testing with the firewall on will only demonstrate that the router blocks connections when the firewall is enabled (which is expected), it won't demonstrate that the connections aren't being blocked by NAT.

People are still running with firewalls on their routers in v6; that's the default basically everywhere.

> Reduces exposure, but does not eliminate it. Agreed though, it is harder to scan ip6 space than ip4, but not impossible

Yup, it's definitely not impossible, but it should be hard enough that the majority of the few unfirewalled Windows machines won't be found, which should weaken botnets rather than strengthen them.

What?! You don’t trust the firewalls software so you are good to trust the NAT software?

Windows firewall? no. iptables? yes. Why? because I know how to configure one and not the other. Also because of things like this:

https://www.enterprisenetworkingplanet.com/security/qa-behin...

Completely true, I wasn't too serious about that, just pointing out that there were unintended benefits.

> But we're at the point in some place where there aren't even enough IPv4 addresses to give to ISP customers even if they're using NAT.

Er, that's a slight exaggeration. It's not uncommon to have 2,000 or more NAT clients behind a single public IP address. 2K * 4B = 8 trillion possible hosts... about 1,000 hosts per living person.

Well, it has limits. You can hide 2000 people behind an address but they might have sporadic connection issues, long held connections such as SSH sessions will be very annoying to support [0] or the users will have to resort to frequent keepalives to make it work regardless, which produces more traffic to destinations you probably don't have in a cache. In some jurisdictions you have to log which customer used what IP (and port) at which point. You can do static assignments but you will have customers (customer homes, e.g. family with perhaps 10 devices or so could be realistic) for which 1000 ports just will not be enough do you will have to dynamically assign spare ports and log those. You might also get more support calls, because the NAT is dropping connections to radically in a peak e.g. a soccer match or whatever. Also the CG-NAT gateway isn't for free and the bigger it needs to be, the pricier it is. Also you might still need to buy some IPv4, perhaps more than you would need to buy if you deployed IPv6 and used it for the connections, where it is possible, taking e.g. almost the full load to Google/ YouTube and Facebook off your gateway.

[0] https://anderstrier.dk/2021/01/11/my-isp-is-killing-my-idle-...

I just found some of my old posts on my old site from 2003, where I complain about NAT and UPNP. It's nearly 20 years, and the problems have stayed the same :(

NAT was the bandaid that became the permanent solution. It was never meant to be permanent, but because of these, so many weird hacks and designs have been made to compensate for it.

Totally silly question, but isn't the NAT-upon-NAT(-upon-NAT) a bit of protection in the sense of additional privacy, and hence a good thing? Doesn't this mean the website cannot directly tie back an incognito user back to a specific IP but instead to just a pool?

For a purely "consumer" internet, yes. But I'd like to think that at some point we move back to a more decentralised model, where addressibility in more contexts than a response to a request become important again. To some degree the NAT issues have even killed a lot of this, making P2P more difficult and unreliable than would be otherwise.

Depends, but really IPv6 and IPv4 seems to be more or less comparable in practice. If you use an app with a login, you are basically screwed anyway. Even on a random website, you can be reliably fingerprinted: https://amiunique.org/

> Doesn't this mean the website cannot directly tie back an incognito user back to a specific IP but instead to just a pool?

Do you have cookies on or off in your web browser settings? If they're enabled then you're being tracking regardless of your IP.

I have my router reboot every night and so I get both a new IPv4 address and a new IPv6 prefix everyday. If I have cookies disabled, then ad folks will have to find other ways to track me beside those two.

Yes, but you could also randomize your IPv6 address constantly too.

Downside of CGNAT and NAT at home is difficulty of getting some services to work, or hosting a server.

It still boils down to everyone gets an identifying number, whether it's an ipv4 address or an ipv6 block.

Any randomized ipv6 address from that block is still as "unique" an identifier as an ipv4 address.

Also, creating/using P2P applications. I needed to look up UDP hole punching and now I wish I hadn't.

out of curiosity, what did you end up going with? I had to host a UDP game server recently and couldn't find a good solution. If only ngrok supported UDP...

It was just research. I'm interested in audio&video over IP, I already had a good grip on HLS, RTP, and a few others, and wanted to dive into WebRTC, RTSP, etc. I don't know but all of this feels like there's a simpler solution (like HLS vs DASH) waiting to be discovered.

DMZ it and firewall up the host.

ISPs log all of the translations and can reverse a (public IP, source port, time stamp) tuple to the user.

> ISPs log all of the translations

On my personal router?

They don't need that much granularity; it is enough to being able to point fingers at someone responsible.

But otherwise, it is possible to enumerate devices behind NAT (obviously, only those that communicate with the world in front of NAT) and assign traffic to them.

No, at their CGNAT layer. They know which customer it came from. They don’t care which device in your house it came from.

"a bit of protection in the sense of additional privacy, and hence a good thing?"

The more dysfunctional is your internet connection, the more prova you youve got

In your first line, I guess you mean to say ipv4 not ipv6.

there's also NAT64, 464XLAT that do ipv6 to ipv4 sorcery

Beyond an ELI5, but correct.

If you are on an IPv4-only or IPv6-only network, you can reach the other through various mechanisms.

A lot of mobile/cell networks are IPv6-only, and so the telcos need a way to allow clients to reach IPv4-only systems.

From a game developer/player perspective: there’s a crazy amount of complexity in connecting between two computers that wouldn’t need to be there in an ipv6 world. See all the NAT, TURN, STUN nonsense that webrtc has to deal with: https://developer.mozilla.org/en-US/docs/Web/API/WebRTC_API/... Today to use webrtc, u need a neutral server to help establish the connection like PeerJS does. This wouldn’t be needed in an ipv6 world. It’s nearly impossible to do pure peer to peer connections today.

Letting machines connect directly to each other in any context, and especially in the context of an online game, is a massive security and privacy risk.

No. Please stop spreading this nonsense that NAT is solution for security and privacy. NAT was solution for getting systems online without increasing address space. It served that well. Now it needs to die.

If you have security issues that is because you failed to configure your firewall properly. Besides Internet was always supposed to work the way IPv6 would allow.

> If you have security issues that is because you failed to configure your firewall properly.

This works for us, but what about average people who have no idea what a firewall is?

Make firewall accessible to average people. Create GUIs, guide them. I mean these are very basics of Internet.

All the major firewalls have GUIs. But try this with some of your non tech family, and tell me how difficult it was. Then imagine this on a global scale. Then imagine how many people are gonna disable said firewall to allow them to play some game or because something asked them too.

I didn't mention NAT. I'm comparing direct connections between peers with using a centralised server.

Enjoy playing an online shooter with extra 200ms of delay, or crappy calls on Teams where massive delay causes you to interrupt each other

Not all games use p2p for matches. Fortnite is the easiest example of one that’s not. Hell, games are not even built the same way anymore and what you think is a server is in reality a cluster of AWS services.

I play counter strike with a delay of around 10 ms with servers that are hosted hundreds of kilometres away.

You mean without? Well you're wrong. It's not a security concern at all. Allowing anyone from the outside to connect to any port they want is a security concern, simply because there's a lot of insecure software people run that doesn't account for malicious connections. However, allowing a user to intentionally let a piece of software listen for outside connections is in no way insecure.

> However, allowing a user to intentionally let a piece of software listen for outside connections is in no way insecure.

It is. Considering the kernel access often given to multiplayer games for anti cheat, and the abysmal attention to security and ability to write secure code by the average application developer, letting Internet randos send arbitrary instructions directly to your machine may not be the best idea.

If the software is executing "arbitrary instructions" from remote sources, then its a malicious piece of software, or so horribly negligent to amount to the same thing. So I don't agree that direct connections are the problem - the badly written software is the problem.

> However, allowing a user to intentionally let a piece of software listen for outside connections is in no way insecure.

Unless you care about security of course. “A user” in your sentence can quite frequently be vulnerable or malicious software.

No, "a user" in my sentence is intended to mean a human. Its very possible to program software to require human approval for connections. Also, if malicious software is already running, then it will be able to make connections to remote servers that it wants, no need for it to passively listen. IPv6 does not make this problem worse.

It’s not nonsense though. The implementation of NAT literally implies a stateful firewall.

I want ipv4 dead as well but to bury your head in the sand and pretend NAT doesn’t offer the protections it does only hurts your argument.

> Besides Internet was always supposed to work the way IPv6 would allow.

Yep, but the real world - where all of the unpatched IoT devices are running - has NAT at basically every home protecting devices from unsolicited connections.

NAT doesn't imply stateful firewall at all. NAT is literally just rewriting IP addresses on incoming/outgoing packets. I could have a single machine behind a middlebox, and the middlebox could just rewrite the IP source/destination of egress/ingress packets, and that would be NAT - and I'd still be able to successfully receive incoming packets from the big bad web. In fact, you can do this without the middlebox using an iptables MASQUERADE rule.

But even then, the added security of a stateful firewall as provided by a router is dubious. You know what else has a "stateful firewall"? Your kernel's TCP/IP stack. It isn't gonna accept random connections from the Internet unless there is an application actively listening to a port and accepting packets. And I trust the Linux/NT/BSD kernel to be more secure with ensuring that than a binary firmware blob from a router manufacturer.

Sorry, you’re very confused about the NAT that is widely deployed. It’s absolutely not 1:1 NAT because that buys effectively nothing from a scaling perspective.

> But even then, the added security of a stateful firewall as provided by a router is dubious. You know what else has a "stateful firewall"? Your kernel's TCP/IP stack. It isn't gonna accept random connections from the Internet unless there is an application actively listening to a port and accepting packets.

That’s the fucking problem. All kinds of vulnerable/misconfigured software just binds to 0.0.0.0:<whatever> and calls it good. My fridge does this, my washer does this, my TV does this. This is the world of IoT.

> And I trust the Linux/NT/BSD kernel to be more secure with ensuring that than a binary firmware blob from a router manufacturer

It’s not, it takes a single API call to have a program start listening because that’s the entire job of the kernel. You have to configure a firewall on top of it to make sure vulnerable software isn’t exposed to the internet.

The kind of NAT that's widely deployed doesn't act as a firewall though. It buys you nothing from a security perspective. It's entirely possible to connect inwards over a NATing router unless there's also an additional firewall configured.

Stateful NAT does imply state tracking, which is a major component needed to implement a stateful firewall, but it is not itself a stateful firewall.

"Beware of appeals to the 'real world'; and to what it supposedly demands. It is always an invitation to leave unchallenged the speaker's tacit assumptions."

Not demands. It’s just the reality of where we are. People who claim everything will improve by allowing p2p connections by default are woefully out of touch with the shit show that is IoT.

The IoT shitshow is never going to improve, if IoT device makers can't be made to stop offloading the cost of their laziness and bad systems thinking onto a network layer that was never intended, and is poorly suited, to be the ne plus ultra of device security.

NAT is not security. If some people need a relay because gamers commit harassment that can used on top of IPv6. Everyone else can connect directly to lower latency.

https://help.steampowered.com/en/faqs/view/1433-AD20-F11D-B7...

You can (and should) still have a firewall that only allows connections to specific ports when using IPv6.

Still you are revealing your IP address to the other parties, which will be more than happy to DoS you to force you to disconnect, exploit 0-days in the game networking code to crash your game or get your private info, know where you are located by IP geolocation...

The idea of P2P in competitive videogames strikes me as absolutely insane

When using NAT, you're revealing the IP address of your router. I don't know about you, but I don't have so many devices running on my home network that would drown out what I'm doing.

With NAT, you can still receive DoS attacks, still have your game networking exploited, and still be geolocated. The only remotely security-related benefit is that instead of your ports being exposed to the wild internet, they're exposed to your router which is more of a side-effect rather than an actual benefit. Its not a reason to not bother having a firewall.

I didn't mention NAT at all. Practically all multiplayer games use a central server. It's impossible to know the IP addresses of your peers.

Nothing about IPv6 prevents continuing to do that when it makes sense (eg to prevent cheaters from dosing their opponents).

"The idea of P2P in competitive videogames strikes me as absolutely insane"

What's insane, is the idea that you want me to use and pay for some crappy AWS server that spies on my data instead of directly connecting to my friend using my own equipment

Well, have fun guessing 2^64 possibilities. YouTube lists even private videos "secured" using 11 BASE64 characters (66 bits in theory, but they seem to use just 64 bits). You can watch Tom Scott explain it: https://www.youtube.com/watch?v=gocwRvLhDf8

CG-NAT doesn't really prevent geolocation. Better services will still pin-point you to the nearest city. There are perhaps easier ways to get your private info or your money - phishing and ransomware seem to be still very popular. Don't have to hack games that only relatively few people have. It is more profitable to attack a bigger market or more wealthy institutions or companies in foreign countries. Also, if you hack the central game server, you will have a lot more victims... Choose your poison.

I guess, there are no games or other software that cannot be audited in high security installations. At home, having a work computer and a game computer (or a VM with GPU pass through or whatever) might be a safer choice in any case independent of IPv4 or IPv6 usage or the quality of your firewall.

When it comes to DoS or geolocation, attackers can just as easily work with your router's address.

Getting rid of NAT doesn't mean that you have to let every machine connect to every other machine. It just means that if you choose to let machines connect, they can do so without their packets needing to be rewritten.

NAT and especially CGNAT has a tendancy to cause problems.

I vividly recall an incident where Jio's CGNAT was dropping idle TCP connections after 10 seconds of idle. (They fixed it, but I don't know if they fixed it for all destinations or only special destinations).

And, of course, I say dropping, rather than closing, because NAT usually doesn't send FINs to close sessions when they're dropped. So you only find out when you try to send more data. Some NATs don't even send RST when you send data on a connection it dropped, so you have to wait for a timeout.

There's also a capacity issue. If you're serving your website via a single IP (common with load balancers), you can only have 65535 connections from each client IP (assuming https on port 443 only, using non default ports is often a non-starter); if the user's ISP is sharing a very limited pool of IPs with a large number of users, it's conceivable that all the connections to your site could fill up.

> if the user's ISP is sharing a very limited pool of IPs with a large number of users, it's conceivable that all the connections to your site could fill up.

Unless each of your users is establishing 600 connections to your site, this isn’t a realistic issue. I don’t know of ISPs that go beyond 1000:1 over-subscription.

I would argue that the main reason is that we are stuck with a centralized Internet with a somewhat large initial step to start a new service. If everyone get IPv6 we are all able to be a first class citizen on the Internet, meaning I can run a webserver or whatever from home. Once I am ready I can move on to a hosting provider with all the extra costs and hassle it comes with. Without IPv6, we are stuck with a two tier system of Internet users. Basically consumers and providers where the step to become a provider is larger than when you can simply start from home with whatever scrap you have in your garage.

IPv6 will also simplify a lot of things (being able to scrap NAT (note, you will still need a firewall) and avoid protocol issues for End-to-End services) but that is just a bonus.

We eventually run out of IPv4 addresses and have to switch anyway

That doesn't work as an explanation.

NAT was developed. CGNAT was developed. HTTP Host differentiation was developed. SNI was developed. STUN was developed. TURN was developed.

The main problem with IPv4 to IPv6 migration is the, frankly, overzealous utopia that "they will migrate anyway". Worse, it's very different that doesn't allow for a simple upgrade option of "IPv4, but longer addresses" because IPv6 is designed to be fundamentally different.

Which is more insulting considering that BGP was seamlessly upgraded from 65536 ASes to more than a million because they only changed the bits in an AS and nothing more. As a router developer, you only need to do the relatively minute changes of extended ASes than the (relative) mess that is IPv6.

And now the internet got worse because you're no longer have technical freedom due to these NAT popping up everywhere. For example, previously if you want to release a video/voice chat app, you just release it and be done with it. Now it won't work like that due to NAT, so you'll have to maintain some STUN and TURN servers just to make your app usable to general public. Think about how many potential innovation got stifled because of added costs and complexity turned some people away before they even started.

I agree. It makes things much harder than it should be.

I wanted to try installing the module into my router so it would pass the WebRTC test for STUN/TURN and perhaps make video even more efficient to allow point to point connection instead of having everything be relayed through a tertiary server, but there was no documentation and having never set up a STUN/TURN server before I couldn't get it to work. So I am dependent on what a third party has set up to facilitate connectivity.

In the old days, you'd just connect to an IP address and that was it.

>NAT was developed. CGNAT was developed. HTTP Host differentiation was developed. SNI was developed. STUN was developed. TURN was developed

And every one of those has introduced its own security flaws and implementation bugs increasing the complexity of applications.

I'm advocating and helping networks to migrate to IPv6, but I'll say that IPv6 was rather a rats-nest design that shot itself in the foot. It's elegant sure, but as Torvalds would say to critics about his frequent use of goto, it's theoretically good but not sound engineering practice.

The thing is we would be addressing in 128-bit IP already if it was done the BGP way: just minimising the necessary changes to implement it. Look at the effectiveness of Itanium versus AMD64 at migrating customers: IPv6 feels like Itanium, while extended ASes work more like AMD64. Whoever designed IPv6 was focused too much at the theoreticals, than looking at what the real world is and then designing around it.

Can you suggest a way of extending v4's address space that works like BGP or AMD64 do?

I've asked a lot of people this question, and the only answers I've ever gotten are either impossible or already implemented by v6. Perhaps you will be the person to finally explain how to do it -- but as far as I can tell, it's just not possible, and I don't think it's fair to criticise v6 for not doing the impossible.

Do we have anything for IPv4 similar to AMD64?

Or are we forever stuck with either going all in to IPv6 or the current hybrid?

Nope, but we're deep in IPv6 anyways so the best course of action is to finish that, just be careful of the pitfalls and not devolving into silliness like this peering fight: https://www.theregister.com/2018/08/28/ipv6_peering_squabble...

IPv6 implementations are everywhere so the extra changes don't really impact networks.

Regardless of monentary charges (and I'm very aware of them, APNIC is even considering grants to help in IPv6 because it's in limbo for so long), even if we activate IPv6 synchronously there's still many, many problems with it, even when you permanently dual-stack it.

One is connectivity. Even theoretically we should have the same network connectivity (so to speak), you can feel the pettiness when Hurricane-Electric and Cogent fought exclusively on IPv6 (https://www.theregister.com/2018/08/28/ipv6_peering_squabble...) I can't believe I'm saying this, but IPv6 isn't free (or at least costs the same as IPv4).

Second is security. Yes, the world is no longer scannable, but consumer routers are a dud when it comes to IPv6 security. This router (https://pierrekim.github.io/blog/2021-01-12-fiberhome-ont-0d...) has serious problems at IPv6 security, and because it's suffix is always at ::0000:0000:0000:0001 and you've eliminate around 18 bits more because of how networks are laid, you've got yourself an IPv6 botnet, which speaking by, how do you block them? A /64 would be a good start, but OVH provides a single address per server (https://www.ovhcloud.com/vps/compare/). That's it. Don't forget that fiasco with MAC addresses (https://datatracker.ietf.org/doc/html/rfc8981), but at least that's solved.

Third, IPv6 should just work as you said. I wish that's true, but embedded devices will trip you up badly. Some don't fallback to IPv4 and instead try and try again using IPv6. This would be a concern while transitioning, but even after that, some would be still stuck because their servers are only operating on IPv4!

Implementing IPv6 is not free lunch if you know the gritty details.

tell me you're bad at math without telling me you're bad at math.

IPv4 address space exhaustion. NAT breaks end-to-end connectivity, making entire classes of applications difficult to deploy, and this is made worse when the NAT gateway is not under a user's control. NAT gateway deployed by ISPs represent a violation of the end-to-end principle and have to be even more complex and stateful than the NAT gateways users often deploy. If you are looking for a specific problem NAT introduces, one good one is that NAT forces users and application developers to use protocols that the NAT gateway properly supports (e.g. TCP, UDP) and to only use those protocols in ways that are "NAT-friendly" (e.g. easy-to-track client-server sessions that are initiated from the private side the gateway) or else to use some NAT-specific protocol as a crutch (UPnP/IGD, STUN, etc.). When both ends of a two-party protocol are behind NAT gateways it can become even more painful and possibly require a third party to be introduced to an application that is otherwise unnecessary for the application.

So at a minimum IPv6 gets us back to the end-to-end principle where Internet peers connect directly to each other without having to negotiate with the network itself. There are a few other advantages, like the fact that users can get large, publicly-routed addresses spaces for their own use (e.g. ISPs giving users a /56 prefix), simplified bridging of private networks (very low probability of collisions in the private IPv6 range), simplified router configuration (link-local addresses are always available so there is no need to assign numbers to every link), and other assorted niche benefits. In theory IPv6 should mean routers become more efficient because it is easier to aggregate prefixes, which should generally benefit Internet users by reducing latency (though at this point there is not much room left for improvement there).