It is very slow to iterate on the hardware side at such a scale for cost reasons. The cost structure is very different from server hardware, where the increase in performance per watt and rack density of newer hardware essentially pays for itself at some point.
But a 10Gbps Port in your top of rack switch is good enough for many many years and multiple server generations, until you make the leap to a higher port speed.
If the chip inside your SDN switch has no understanding of IPv6, you could only tunnel such packets in software at substantial performance costs. So it takes ages until the whole physical network is using new enough gear.
example.us-east-1.elb.amazonaws.com (A Record)
ipv6.example.us-east-1.elb.amazonaws.com (AAAA Record)
dualstack.example.us-east-1.elb.amazonaws.com (A or AAAA Record)
These records will only be displayed for ELBs in classic despite existing and being resolvable for VPC ELBs. The twist is that there is no way to add security groups with ipv6 addresses in VPC so its not possible to receive requests.
Amusingly this means ELBs in classic have an advantage on VPC when it comes to ipv6.
VPC gives you your own entire rfc1918 space per region, which tho can be a pain to deal with overlap if you have site-to-site VPN, is enough for basically anyone.
This pain point is one of the reasons we've built Wormhole.
[prhodes@doctorfeelgood ~]$ ping6 -n www.fogbeam.com
PING www.fogbeam.com(2600:3c02::f03c:91ff:fe84:7b1d) 56 data bytes
64 bytes from 2600:3c02::f03c:91ff:fe84:7b1d: icmp_seq=1 ttl=55 time=31.7 ms
64 bytes from 2600:3c02::f03c:91ff:fe84:7b1d: icmp_seq=2 ttl=55 time=39.8 ms
64 bytes from 2600:3c02::f03c:91ff:fe84:7b1d: icmp_seq=3 ttl=55 time=38.6 ms
^CJust looking at this makes me feel for network operators that have to troubleshoot layer 3 addresses directly and without DNS. There needs to be a better approach - like an intelligent clipboard.
http://networkrecipes.blogspot.com/2014/08/ipv6-shortcuts.ht...