Introduction

What is IPv6?

IPv6 is the next generation of the Internet Protocol after IPv4, first formalized in RFC 1883 published in 1995

	IPv4	IPv6
Address length	32 bits	128 bits
Number of unique addresses	4.3 billion	340 billion billion billion billion
Address notation	Decimal, with a `.` between bytes `192.0.2.2`	Hexadecimal, with a `:` between two-byte words `2001:0db8:0000:0000:0000:0000:0000:0123` May be abbreviated/compressed¹ `2001:db8::123`
Subnet notation	Network ID and subnet mask `192.0.2.0`, `255.255.255.0` CIDR notation `192.0.2.0/24`	Prefix/prefix length `2001:db8::/64`
Subnetting	Variable Length Subnet Mask (VLSM): Subnet size is based on expected number of clients	Standard subnet size: /64
Fragmentation	Packets may be fragmented in-flight by routers	Packets must be sized correctly by the sender
Number of header fields	12	8
Header Length	20 bytes, variable	40 bytes, plus optional extensions
Header checksum	Yes	No
Host configuration	Static, DHCP, or APIPA (link-local)	Static, DHCP, or SLAAC, and link-local
IP to MAC address resolution	Address Resolution Protocol (ARP) (broadcast)	Neighbor Discovery Protocol (NDP) (multicast)

Address exhaustion
- Difficult/expensive to get IPv4 addresses
- Need for NAT
Internet routing table disaggregation
- Large blocks are being split up and sold off as demand for IPv4 increases
Forwarding inefficiencies
- Packet checksum must be recalculated every time the packet is forwaded by a routers due to TTL decrement
- Header length is variable, making it more complicated to identify the fields
Reliance on broadcast for ARP and DHCP
Server infrastructure required for host configuration

Virtually unlimited address space and subnet size
- IPv6 addresses are free
- No need for NAT! Native communication between unique IPv6 addresses
No reason not to assign aggregable prefixes
Forwarding efficiency
- No packet checksum, and no calculation for the router to perform
- Header length is fixed, so fields are always in the same place
MAC address discovery and host configuration use multicast instead of broadcast
Very simple host autoconfiguration

IPv6 is becoming the dominant protocol - adoption in the US is approaching 50%
- Google IPv6 statistics
  - May 2020: 32%
  - May 2021: 35%
  - May 2022: 40%
  - May 2023: 43%
- IPv4 will become the minority of traffic
Providers are prioritizing IPv6 deployements
- New investments into IPv6, less in IPv4
- Facebook - 100% native IPv6 servers, IPv4 services through proxies/load balancers
- T-Mobile - 100% native IPv6 cellular network, IPv4 services via 464XLAT translation
- Most major content providers support IPv6 (Google, Cloudflare, Akamai, Apple, Microsoft, Facebook)
IPv6 is marginally faster
- Less routing overhead
- Easier forwarding
No NAT required
- NAT breaks or cripples many peer-to-peer protocols
  - Voice/video calling
  - Gaming
  - VPN
- NAT makes network administration more complicated by obscuring addresses
IPv6 is supported by all major carriers
Running dual-stack reduces the impact of a problem with one protocol
- "Happy Eyeballs" (RFC 6555) is a mechanism used by the client OS to choose IPv6 or IPv4 for a particular connection
- If IPv6 is working, hosts will choose IPv6
- If IPv6 is broken, hosts will pick IPv4
IPv6 is very easy to deploy

Address compression rules: a single run of sequential zeroes can be replaced with a :: and leading zeroes may be dropped in each address chunk ↩