[HeckFeck]

> They can't use protocols, they can only consume third party services over HTTP/S for the most part.

Your comment brings a whole new angle to IPv6 I hadn't before considered. Having each packet routable straight to a TCP/UDP port number eliminates the complexities of NAT. Since the {hard,soft}ware that handles NAT wouldn't be needed, perhaps a shift to IPv6 could also give throughput gains.

[wmf]

Since the {hard,soft}ware that handles NAT wouldn't be needed, perhaps a shift to IPv6 could also give throughput gains.

It does, although barely noticeable. More noticeable is the cost to ISPs for carrier-grade NAT equipment.

[ATsch]

IPv6 is already faster on it's own, because the packet header was modified to be more easily routed in hardware.

[snuxoll]

We also took advantage of the stupid huge address space to make routing tables smaller, it's not necessary to have every /24 (or /64 in IPv6 equivalents) in your tables because we can assign big blocks and stop dealing with fragmented networks all over the place.

[gerdesj]

Yes, provided your ISP allocates you a P2P link as well as your prefix. Once you get used to breaking up your prefix into subnets it all starts to work quite nicely.

I have even found that real world IPv6 addresses are not quite as bad as they look. We have all seen the auto-generated ones and they look awful but you don't actually use those as such. For example you are allocated a /48 prefix: 2001:db8:1234:: in general you might think of the ISP as 2001:db8:: and your site as 2001:db8:1234::. Your first subnet might be say 000a or simply "a" for VLAN 10 because you decide not to allocate IPv6 to your default VLAN 1 and start with your current PC VLAN which is 10.123.10/24. In IPv4 you have 10.123.10.1 as the default gateway (VRRP) with .2 and .3 for the physical routers. Hence your PC VLAN routers get given 2001:db8:1234:a::2 and 3 and ::1 for the VRRP address. You can play games with 1337 addresses using 0-9a-f eg face:b00c sigh.