Slashdot Mirror
IPv6 Achieves 50% Reach On Major US Carriers (worldipv6launch.org)
Long-time Slashdot reader dyork brings new from The Internet Society: IPv6 deployment hit a milestone this month related to the four major US providers (Verizon Wireless, T-Mobile USA, Sprint, AT&T): "IPv6 is the dominant protocol for traffic from those mobile networks to major IPv6-capable content providers."
A graph on their "World IPv6 Launch" site shows those carriers are now delivering close to 55% of their traffic over IPv6 to major IPv6-capable content providers -- up from just 37.59% in December. "This is really remarkable progress in the four years since World IPv6 Launch in 2012, and the growth of IPv6 deployment in 2016 is showing no signs of abating." In fact, the NTIA is now requesting feedback from organizations that have already implemented IPv6, noting that while we've used up all the 4.3 billion IPv4 addresses, IPv6 offers 340 undecillion IP addresses -- that is, 340 followed by 36 digits.
A graph on their "World IPv6 Launch" site shows those carriers are now delivering close to 55% of their traffic over IPv6 to major IPv6-capable content providers -- up from just 37.59% in December. "This is really remarkable progress in the four years since World IPv6 Launch in 2012, and the growth of IPv6 deployment in 2016 is showing no signs of abating." In fact, the NTIA is now requesting feedback from organizations that have already implemented IPv6, noting that while we've used up all the 4.3 billion IPv4 addresses, IPv6 offers 340 undecillion IP addresses -- that is, 340 followed by 36 digits.
IPv6 = 256^6 = ... 340x10^36 ???
Not sure if bad attempt at joke or not, but in case it isn't: the 6 in IPv6 isn't the number of octets used in the addresses, it's a version number. IPv6 uses 128 bit addresses, and 2^128 = 3.4 * 10^38.
You recall xkcd 865.
No. PAT involves using port numbers to supplement IPv4 addresses, so that a hybrid static-dynamic NAT in IPv4 can get a 1:1 mapping b/w the local IP addresses and the external IP address coupled w/ the port number. That's one of the things that IPv6 eliminates, but it also removes things like load balancing or address isolation.
NPT - Network Prefix Translation - is different. It keeps the Interface ID unchanged - the part of the address that's not part of the network address - is not touched. Instead, a public unicast address is converted into a site local address, w/o touching port numbers. What you have is a 1:1 relationship b/w local and routable addresses. In PAT, what you have is a 1:1 relationship b/w a routable address coupled w/ one of 65536 port numbers to a local address. Which is a mess, b'cos if one needs ports for anything else (like map segments in a mapping application), one has to keep tabs on the ports used for PAT vs the ones used for the applications that need them
While I do think IPv6 addresses are wasted, I don't think the wastage happens in the global prefix. Rather, it happens in the Interface ID area. 64 bits is way overkill for an interface ID, since no subnet will have anywhere close to even 4 billion nodes, much less 2^64. In the meantime, in the global prefix side, things are squished w/o getting into a hierarchical routing, which would have been a real godsend, had it been implemented.
Yeah, my example probably shifted things, due to my use of documentation's 2001:db8::/32, and your scenario would be the more likely one. Still, I believe that the upper 4 words should have been strictly global prefixes, not including subneting: it should have been used to hierarchically route from IANA to RIR to country to ISP to organization/families/individuals. At a gateway, allow for either a /64 or a /96 (w/ 4 billion subnets), depending on the need. So if a home router needs 2 network addresses for 2 SSIDs, either get 2 /64s or subnet 1 /64 into /96s.
The reason that is given for the 64 bit interface ID is auto-configuration, but that's a lame excuse. First of all, even w/ 64 bits, it's unlikely, but not guaranteed that there won't be an address conflict, and at any rate, there is DAD to resolve that in IPv6. Then, using things like MAC addresses or IMEA numbers to obtain these addresses creates a potential for spoofing agents to deduce a target address, assuming that a network wants those things either hidden, or difficult to find w/o initiating from the user's end. 32 bits would easily have been enough for any subnet - even in the most crowded spot in Guangzhou, I doubt that there would be anything close to 4 billion devices that would be under a subnet, and even if there was, that network would grind to a halt w/o redundant APs, repeaters and other signal enhancing agents. So the IETF could have designed IPv6 to have the first 64 bits strictly the global prefix, then either have a 16:48 or 32:32 split b/w subnets and nodes.