Hacker News Reader

Comment by DaiPlusPlus | Hacker News Reader

IPv6's has a fixed subnet size of 64-bits[1]. A router needs its own subnet, ergo everyone gets their own 64-bit address range. ISPs cannot give each customer a single IPv6 address because that would require each customer to have a single device connected directly to their router with no router in the CPE - basically like how dial-up used to work. There's no demand for a service like that.

[1] See https://www.ripe.net/about-us/press-centre/understanding-ip-... . I note that there is a lot of misinformation and badly-informed blog posts in the top search results for "ipv6 subnet size" like this one, ugh: https://www.crucial.com.au/blog/2011/04/15/ipv6-subnet-cheat... - which is wrong because IPv6 simply doesn't have 128-bit-sized subnet option (whereas you can have a 32-bit sized subnet in IPv4, which is how most residential ISPs/CPEs are configured).

View on HN

It is not true that an ISP cannot give each customer just a single IPv6 address. The Linux box I use as a router gets a single (i.e., /128) address from Spectrum Cable on its upstream interface, and it has to explicitly ask for a separate /64 prefix to advertise on the local interface (this is called prefix delegation, PD).

So it's certainly possible for an ISP to give a customer just one IPv6 address. I'm not sure, but I think there is no rule that prefixes obtained with PD must be /64's. They could in principle give you a /126 or /127 to limit you to just a few hosts on your LAN, and charge to give you anything better. This would break stateless autoconfiguration but it would still work with DHCPv6.

Your best workaround is still www.tunnelbroker.com, HE's free IPv6 tunnel service.

Thank you for correcting my misunderstanding of IPv6! :)

> The Linux box I use as a router gets a single (i.e., /128) address from Spectrum Cable on its upstream interface, and it has to explicitly ask for a separate /64 prefix to advertise on the local interface

Ask for a /56! Spectrum in Los Angeles hands out /56s, and I understand many other residential IPv6 ISPs do as well. You get your own 8 bits worth of subnets on top of /64 for easy SLAAC address assignment. This works "for free" not only because the IPv6 address space is large enough to do this, but more importantly because IPv6 does NOT have any fixed subnet sizes - routing is hierarchical and subnet assignment is transparent. /64 is just the size used by SLAAC address auto-configuration, another bit of IPv6 magic.

Phil,

Thank you so very much for NOS, which was my very first introduction to IP over packet and what initially got me interested in amateur radio (going on 30 years ago!) and is, in part, what led to me initially getting licensed and drawn into the world of networking at a young age.

73,

N9WWV

P.S. Welcome to HN!

RIPE just has a rule about only having ISPs hand out those /64s there is nothing to prevent someone internally to then split that /64 into smaller chunks to hand out to internal clients. Thats the would point of a subnet!

I'm not sure if that's well-supported by IPv6 though. Documents I've read do agree that it's possible to have smaller subnets in IPv6 (e.g. https://docs.netgate.com/pfsense/en/latest/network/ipv6/subn... and https://serverfault.com/questions/426183/how-does-ipv6-subne... ) but that the pain and hassle required to do-so isn't worth it.

Wait, so does that mean that every customer would have 18.45 QUINTILLION addresses to use?

God, I knew the IPv6 space was insanely large, but Jesus h, man.

2001:: is a 16 bit prefix in which every single customer could get a /48, i.e. you could allocate 65k subnets of 18.45 quint addresses on your network and give one of those to everyone with an ipv4 address and not run out, and that's still only 1/65536th of the total address space.

128 bits is a lot.

(the /64 as the physical subnet allows using hardware MAC addresses and RNG temporary/private addresses instead of needing DHCP/etc).

Reply on news.ycombinator.com