more tech thoughts

[kirkcudbright]

As previously noted, I released new software on Monday last week. The press release that was supposed to go out at the same time...still hasn't.

I gave a tech talk Thursday last week. The video that was supposed to be available by Friday...still isn't.

I have to admit I'm a bit conflicted about the technology. If the ISPs had been doing their job, they would have turned on IPv6 years ago, and we'd already be living in a dual-stack world. But there are costs associated with doing that (upgrading routers, upgrading home gateways, upgrading debugging and network management skills), and up til now, there's been no cost associated with not doing it. So it was pretty inevitable that we'd come to the run-in-circles-scream-and-shout stage before anything got done.

In my ideal world (which might still come to pass), we wouldn't need any kind of address-sharing technology (beyond the NAT we've already got. The network core (carrier networks, backbones, etc) could all go to IPv6 immediately. Servers should already be dual stack capable, if their providers would only provision them with v6. Many many end user devices could go dual stack or v6-only without users noticing. OTOH, my printer will never be v6-capable, but I'm not putting it on the global internet.

If it comes to the point where the carriers have to implement IPv4 address sharing, it will make everyone's life that much more complicated - users and providers both. We'll be ready for that eventuality, but it feels like we're doing the wrong thing for all the right reasons.

Comments

[nw1.livejournal.com]

Speaking as someone intimately involved with the IPv6 initiative here at Yahoo!, the server side isn't quite as simple as that. Some server code needs to be rewritten. (examples: code that stores IP addresses as integers, code that depends on DNS round-robins working in a particular way, etc.) Even code that doesn't itself interact directly with the network, will have elements in it's toolchain or library set which does, and which will need to be upgraded (necessitating a full test cycle at the very minimum).

Dual-stack is clearly the way forward though, you're right about that.

[kirkcudbright.livejournal.com]

I should have been clearer. The server platforms (*nix, *bsd) are already dual-stack. The major server apps (apache, exim, postfix) are v6-capable. Clearly there are many other apps that need to be converted, and even probably some plugins to mail/web servers. There's work to be done, but it's work we know how to do.

PS. Just assume we had the discussion about how relying on DNS result ordering is a really bad idea.

[nw1.livejournal.com]

Agreed, the work that needs to be done is work we know how to do. The difficulty, is, as always, convincing the people who have the authority to make sure the work needed gets the appropriate priority, and makes it into the schedule.

P.S. Relying on DNS result ordering is indeed a really bad idea, and yet a surprising number of developers think it's a good idea.

[kirkcudbright.livejournal.com]

It's Y2K all over again!

[chocorua.livejournal.com]

We (mostly me) converted our network analysis tool to be address-agnostic several years ago. There are a bunch of things that the C/Unix/database community (both commercial and open-source) could have done to accelerate this, in the form of data types, APIs, display widgets etc. When I looked for them, they weren't, so we had to roll our own. As the pressure starts to build, perhaps we'll get our collective eye on the ball.