The lived reality for most traffic by packetflow is that it goes from you, through your ISP, to a CDN close to your ISP logistically speaking. It has almost zero global routing impact, and global addressing impact is low. It might as well be six as not. It might as well be RFC1918 all the way as not. It's not really going to alter much. If you are used to how things worked 10 or 15 years ago, the moment of change has passed, and "internet" is the same but kinda different. packets? same. Paths? Same out from you but goes to different places.
The role of DNS in steering has overtaken the role of BGP routing.
The lived cost, which they discuss, is that on-link device discovery for people who want direct contact becomes harder, because you're ships passing in the night using ARP and an analogous function in IPv6 but they don't intermix. So, somebody has to either mediate (proxy) or be dualstack.
For everything else, in most cases it's already doing NAT agile protocol work up the stack, and already using mediated conversation. Not that true end-to-end doesn't exist, but it's just not as common as people like to think.
I think Nick, Ondrej and Jen do a good job of documenting things. There's more typos than I expected in an 04 but they'll get sorted out.
(none of this means it isn't still a headache for a network operator, or an applications designer, or anyone else. Just that it's better not to think you know what it looks like if you are (like me) an old timer, because it's a different ship now)
The lived reality for most traffic by packetflow is that it goes from you, through your ISP, to a CDN close to your ISP logistically speaking. It has almost zero global routing impact, and global addressing impact is low. It might as well be six as not. It might as well be RFC1918 all the way as not. It's not really going to alter much. If you are used to how things worked 10 or 15 years ago, the moment of change has passed, and "internet" is the same but kinda different. packets? same. Paths? Same out from you but goes to different places.
The role of DNS in steering has overtaken the role of BGP routing.
The lived cost, which they discuss, is that on-link device discovery for people who want direct contact becomes harder, because you're ships passing in the night using ARP and an analogous function in IPv6 but they don't intermix. So, somebody has to either mediate (proxy) or be dualstack.
For everything else, in most cases it's already doing NAT agile protocol work up the stack, and already using mediated conversation. Not that true end-to-end doesn't exist, but it's just not as common as people like to think.
I think Nick, Ondrej and Jen do a good job of documenting things. There's more typos than I expected in an 04 but they'll get sorted out.
(none of this means it isn't still a headache for a network operator, or an applications designer, or anyone else. Just that it's better not to think you know what it looks like if you are (like me) an old timer, because it's a different ship now)