Posted Jan 26, 2011 13:11 UTC (Wed) by dwmw2
Parent article: LCA: IP address exhaustion and the end of the open net
Geoff talked a lot about IPv6 being 'unreliable', and spoke of the 'white screen of death' which you get in your web browser when trying to connect to a dual-stack server via IPv6 first, when you don't actually have working IPv6 connectivity. As he says, it does indeed take quite a while for the connection to fail after a number of SYN packets don't get responses, and for the stack to fall back to using Legacy IP.
But there are two important things to note about this failure mode that he made so much of:
- It's rare. According to Wikipedia's testing it's less than half a percent of clients that would screw up by trying IPv6 if an AAAA record was advertised, when they don't really have IPv6 connectivity
- It's a local problem on the client side. It's just the same as having a rogue DHCP server, giving you a false IP address and routes and thus breaking your Legacy IP connectivity. As soon as people start noticing their local problem because sites like Google, Facebook etc. just go ahead and publish AAAA records for their main hostnames instead of wimping out and using separate hostnames, even that 0.40% will be able to fix things fairly quickly.
As Legacy IP becomes afflicted with more and more levels of NAT, its
reliability will fall; as the reliability of IPv6 increases because people have fixed their rogue route advertisements. The dystopic picture that Geoff paints of "Enable IPv6, see the white screen of death"
will become even less realistic than it is today.
In the last decade or so that I've been running IPv6, I've lost count of the number of times that it has saved my connectivity.
Even at big conferences with only a single level of NAT, I've seen port-space exhaustion (and NAT table exhaustion) cause connectivity problems for the Legacy IP users, while those using IPv6 have been able to connect just fine.
to post comments)