4 ms·

Tweaks to IPv4 could free up millions of addresses

11 points by petercooper 4y ago

mc4ndr3 4y ago
Treat IPv6 as an extension of IPv4, not a separate network space. Deprecate software that doesn't support IPv6. Stop DHCP servers from providing IPv4 by default. Then we can enjoy a world with mass 6 adoption, almost no hosts bound to 4 addresses, and billions of available 4 addresses.
- sieabahlpark 4y ago
  What an exceptionally wasteful idea.
- p1mrx 4y ago
  > Treat IPv6 as an extension of IPv4, not a separate network space. We could have rolled out 6to4 with no anycast gateways, and the entire internet numbered from 2002:<ipv4>::/48. Every node could have IPv6, using only software updates and no cooperation from ISPs or RIRs. That approach would have had some interesting positives and negatives, but the time to discuss it was >20 years ago.
  - welterde 4y ago
    Except every node still needs an IPv4 address for that to work since it doesn't play nice with NAT. It also doesn't solve the issue of not being able to talk to IPv4-only hosts, since the issue is not the backbone or ISP anymore, but people building their stuff only with IPv4 in mind and don't care enough to add support for anything else. It would have gained us very little and we would be in an even worse situation now (in some countries we are at > 50% IPv6 in the real world now!)
t-3 4y ago
Just let it die already! IPv6 isn't perfect, but it's a hell of a lot better than v4. The scarcity of addresses is a problem that serves no purpose but enabling middle-men. At this rate, they'll probably have to be replaced together, once wireless technologies improve to the point that demand for mesh networking and multipath is great enough to demand new layer 3 protocols.
exabrial 4y ago
Odd man out: I'm not comfortable with ipv6 in a world of Facebook and Google, and oddly enough they're the ones pushing the hardest for it. And before you lecture me how privacy extensions and address blocks, I think the beauty of ipv4 is NAT works and is the default, providing plausible deniability as a core feature. If ipv6 did the same, I'd be onboard. Ipv6 drops so many legacy features of ipv4 and provides flow identification as part of the protocol. It's a damned shame in my opinion, as I do think it was designed "Better".
- p1mrx 4y ago
  Google and Facebook are pushing IPv6 primarily because those companies attract nerds who understand the problem. IPv4 NAT does provide some degree of privacy, but law enforcement can still identify you if the TCP source port is logged. Is this incremental privacy benefit really worth hampering the growth of the internet forever? The goal of IP is to route packets from A to B, which fundamentally requires knowing the locations of A and B. Privacy and efficiency are contradictory goals, and it's much easier to build privacy on top of an efficient network (e.g. using proxies and VPNs) than the other way around.
- welterde 4y ago
  > I'm not comfortable with ipv6 in a world of Facebook and Google Instead you will get a world that's controlled by Amazon, Google and Azure, since they are hoarding as much IPv4 space as they can get their hands on (amazon alone has over 2 billion $ worth of IPv4 space). For startups it gets more and more expensive to acquire the necessary ip space. > providing plausible deniability as a core feature. Not really. In the non-carrier grade NAT (CGN) case (where you just do NAT at home) you still have a 1:1 mapping to your home network and the situation is exactly as good/bad as it is with IPv6. But with CGN the ISPs now have to keep logs on all connections being made for abuse tracking.. so much better (depending on the CGN implementation of course). > provides flow identification as part of the protocol. That's a good thing though and not any worse than before. Routers have to load balance traffic across multiple different paths and they do that based on flows (since load balancing one flow across multiple paths can cause reordering, etc.). And since src_host and dst_host is not enough entropy for that, typically routers also look at the TCP/UDP source and destination port to identify (with funny results when you use non-tcp/udp protocols or have different hashing algorithm between different routers). With the flow label the router just has to look at the IPv6 header and not concern itself whether it's a tcp or udp packet.

11 points by petercooper 4y ago

Treat IPv6 as an extension of IPv4, not a separate network space. Deprecate software that doesn't support IPv6. Stop DHCP servers from providing IPv4 by default. Then we can enjoy a world with mass 6 adoption, almost no hosts bound to 4 addresses, and billions of available 4 addresses.

sieabahlpark 4y ago
What an exceptionally wasteful idea.
p1mrx 4y ago
> Treat IPv6 as an extension of IPv4, not a separate network space. We could have rolled out 6to4 with no anycast gateways, and the entire internet numbered from 2002:<ipv4>::/48. Every node could have IPv6, using only software updates and no cooperation from ISPs or RIRs. That approach would have had some interesting positives and negatives, but the time to discuss it was >20 years ago.
- welterde 4y ago
  Except every node still needs an IPv4 address for that to work since it doesn't play nice with NAT. It also doesn't solve the issue of not being able to talk to IPv4-only hosts, since the issue is not the backbone or ISP anymore, but people building their stuff only with IPv4 in mind and don't care enough to add support for anything else. It would have gained us very little and we would be in an even worse situation now (in some countries we are at > 50% IPv6 in the real world now!)

t-3 4y ago

Just let it die already! IPv6 isn't perfect, but it's a hell of a lot better than v4. The scarcity of addresses is a problem that serves no purpose but enabling middle-men. At this rate, they'll probably have to be replaced together, once wireless technologies improve to the point that demand for mesh networking and multipath is great enough to demand new layer 3 protocols.

exabrial 4y ago

Odd man out: I'm not comfortable with ipv6 in a world of Facebook and Google, and oddly enough they're the ones pushing the hardest for it. And before you lecture me how privacy extensions and address blocks, I think the beauty of ipv4 is NAT works and is the default, providing plausible deniability as a core feature. If ipv6 did the same, I'd be onboard. Ipv6 drops so many legacy features of ipv4 and provides flow identification as part of the protocol. It's a damned shame in my opinion, as I do think it was designed "Better".

p1mrx 4y ago
Google and Facebook are pushing IPv6 primarily because those companies attract nerds who understand the problem. IPv4 NAT does provide some degree of privacy, but law enforcement can still identify you if the TCP source port is logged. Is this incremental privacy benefit really worth hampering the growth of the internet forever? The goal of IP is to route packets from A to B, which fundamentally requires knowing the locations of A and B. Privacy and efficiency are contradictory goals, and it's much easier to build privacy on top of an efficient network (e.g. using proxies and VPNs) than the other way around.
welterde 4y ago
> I'm not comfortable with ipv6 in a world of Facebook and Google Instead you will get a world that's controlled by Amazon, Google and Azure, since they are hoarding as much IPv4 space as they can get their hands on (amazon alone has over 2 billion $ worth of IPv4 space). For startups it gets more and more expensive to acquire the necessary ip space. > providing plausible deniability as a core feature. Not really. In the non-carrier grade NAT (CGN) case (where you just do NAT at home) you still have a 1:1 mapping to your home network and the situation is exactly as good/bad as it is with IPv6. But with CGN the ISPs now have to keep logs on all connections being made for abuse tracking.. so much better (depending on the CGN implementation of course). > provides flow identification as part of the protocol. That's a good thing though and not any worse than before. Routers have to load balance traffic across multiple different paths and they do that based on flows (since load balancing one flow across multiple paths can cause reordering, etc.). And since src_host and dst_host is not enough entropy for that, typically routers also look at the TCP/UDP source and destination port to identify (with funny results when you use non-tcp/udp protocols or have different hashing algorithm between different routers). With the flow label the router just has to look at the IPv6 header and not concern itself whether it's a tcp or udp packet.