Hacker News Reader

Comment by 3h49ry7zg | Hacker News Reader

I'm quite confused why this is getting attention. It's not hugely difficult to enable a public IPv6 address on a server alongside the IPv4 address. The advantage of doing so? If one runs a serious website, there are many more back-end servers that handle database requests etc. than there are user-facing webservers, and all of these back-end servers can use IPv6 to talk to each other and the webservers. If IPv6 addresses are cheaper and easier to get, then it's a win for the company. They can upgrade each internal client-server pair to IPv6 in unison since they're a single organization.

View on HN

As if the number of servers using IPv6 mattered the slightest. It does not. What matters is the number of IPv6 servers that can talk to all clients. That means IPv4 clients as well (they are the majority, after all).

Being connected to the internet means being able to talk to every single machine on that network. If a computer can't talk to IPv4 computers, then it is by definition not connected to the internet.

We can't split the network. If we have to do it to switch to IPv6, then it just won't happen. Ever. To avoid splitting the network IPv4 and IPv6 machine have to be able to talk together.

So, as long as IPv6 and IPv4 computers can't talk to each other, we will be limited to the back end servers you are talking about. Until we go beyond that point, IPv6 will stay relatively useless.

As an ordinary user, I can abandon ipv4 as soon as the services that I use are on ipv6. I don't need to be able to connect to every one of the 500 million computers on the network via ipv6 - I only need to be able to connect to the servers and services that I actually use - Google, Amazon, YCombinator, etc.

Worst case, I'd need to connect to all the computers in my P2P network - a few million, or all the computers in my bot net - a few million more. And in both of those cases, the client software either is or will be smart enough to keep me connected via a proxy of some sort.

For example - To get to Google without ipv4, I don't need to connect to every one of their millions of servers via IPV6, I only need to get a quad A from my DNS and connect to their external facing load balancers via ipv6. As soon as I can do that, I don't need ipv4 to use Google. For all I care they could be running Arcnet on their back end servers.

Another consideration is the prevalence of NAT and proxies in the corporate world. Corp users are already NAT'd, Proxied, Content Filtered anyway, so upgrading the the corporate web proxy to do 4 to 6 translation puts the entire corp desktop on the 6 network in one shot, and isn't a major re-think of Corp networking.

That's a silly argument - any site which uses a non-trivial number of backend servers will have an internal network using RFC1918 v4 addresses, and there's no shortage of those (djb covers this).

As someone who runs such a site, I'd only think about upgrading the internal networks to v6 long after the public internet has gone to v6.

Suppose you are using a cloud hosting service, or multiple such services, and don't have the luxury of an internal network? Then your network IS the public internet.

I've heard that for very large service providers, it is possible to run out of 1918's. IIRC Comcast did a presentation where they figured they needed 100 million IP addresses for a nation wide rollout of their triple play service. They appear to be highly motivated to use 6, as the alternative - partitioned 1918's would be a nightmare.

I agree, though, that exposing your services to the Internet via both 6 and 4 and leaving the backend internal data center at 4 is probably the way to go for the next decade or so. Heck - we are still running DecNet on production servers.

Reply on news.ycombinator.com