An interactive lab for the 128-bit internet
IPv6 Workbench
Learn IPv6 by taking it apart. Type addresses and watch them decode, carve up subnets, assemble packet headers, and run Neighbor Discovery frame by frame on a simulated LAN. Every tool on this page runs in your browser.
- Addresses in IPv6
- 2128340,
282, 366, 920, 938, 463, 463, 374, 607, 431, 768, 211, 456 - Addresses in one LAN (/64)
- 18.4 quintillion4.3 billion times the entire IPv4 internet, in a single subnet.
- Google users on IPv6
- 50.1%on 28 March 2026, the first day above half.
- Base header
- 40 bytesAlways the same size. No header checksum, and no broadcast anywhere in the protocol.
Why IPv6 exists
IPv4 has 4,294,967,296 addresses. The regional registries handed out their last free blocks between 2011 and 2020.
An IPv4 address is 32 bits long, so there are 232 of them, about 4.3 billion. Large slices were never usable on the public internet (private ranges, multicast, reserved space), and most people now carry several connected devices. The free pool ran out region by region.
The stopgap was NAT: many devices share one public address, and many carriers add a second layer called carrier-grade NAT. It works for browsing, but it breaks end-to-end reachability, complicates peer-to-peer apps, calls and games, and makes it hard to trace abuse to one customer.
IPv6 was designed in the mid-1990s (RFC 1883 in 1995, RFC 2460 in 1998) and became a full Internet Standard as RFC 8200 in 2017. Its addresses are 128 bits, so every LAN gets 264 addresses. It also simplified the packet header, made autoconfiguration standard, and replaced broadcast with multicast.
How big is 128 bits?
Bars use a log scale: bar length is the number of bits (log₂ of the count), so each extra bit doubles the count. Select a bar for the exact number.
Thirty years: IPv4 runs out, IPv6 arrives
IPv4 and IPv6 side by side
| Topic | IPv4 | IPv6 |
|---|---|---|
| Address size | 32 bits, about 4.3 billion | 128 bits, about 3.4 × 1038 |
| Written as | 192.0.2.1 | 2001:db8::1 |
| Header | 20–60 bytes, variable, with options | 40 bytes fixed; options move to extension headers |
| Header checksum | Yes, recomputed at every hop | None; link and transport layers check integrity |
| Fragmentation | Routers and hosts | Only the sending host; routers send Packet Too Big |
| Minimum link MTU | 68 bytes (hosts must accept 576) | 1,280 bytes |
| Finding a neighbor's MAC | ARP, by broadcast | Neighbor Discovery, by ICMPv6 multicast |
| Broadcast | Yes | None; multicast groups such as ff02::1 instead |
| Automatic addressing | DHCP | SLAAC, DHCPv6, or both |
| Addresses per interface | Usually one | Several: link-local, global, temporary, sometimes ULA |
| Private addressing | RFC 1918 ranges plus NAT | Unique local fd00::/8; NAT not needed |
| DNS | A records, reverse in in-addr.arpa | AAAA records, reverse in ip6.arpa |
Reading and writing addresses
Eight groups of four hex digits, plus two shortcuts that make them human-sized.
An address is 128 bits. We write it as eight 16-bit groups, often called hextets, in hexadecimal, separated by colons: 2001:0db8:0000:0000:0000:ff00:0042:8329.
- Drop leading zeros inside any group:
0db8becomesdb8,0042becomes42,0000becomes0. - Replace one run of consecutive all-zero groups with
::. Only once per address, or nobody could tell how many groups each::hides.
RFC 5952 defines one canonical form so that logs and databases agree: lowercase, as short as possible, :: on the longest run of zero groups (the leftmost one if two runs tie), and never :: for a single zero group.
A /n suffix is the prefix length: how many leading bits name the network. A normal LAN is a /64, leaving 64 bits for the interface ID. In URLs, put the address in brackets so its colons aren't read as a port: http://[2001:db8::1]:8080/. Link-local addresses need a zone ID naming the interface: fe80::1%eth0.
Sandbox · Address explorer
Compression, step by step
Follows whatever address is in the explorer.
Practice drill
Address types and scopes
Every address has a type (unicast, multicast or anycast) and a scope (how far it may travel). The first few bits tell you which.
- Unicast goes to one interface. The everyday kinds are global unicast (
2000::/3, routable on the internet), link-local (fe80::/10, on every interface, never forwarded by routers) and unique local (fc00::/7, for private networks; in practicefd00::/8plus a random 40-bit Global ID). - Multicast (
ff00::/8) goes to every member of a group. IPv6 has no broadcast: jobs that used broadcast in IPv4 now use groups such asff02::1(all nodes on the link) andff02::2(all routers). - Anycast is one unicast address configured on several interfaces; routing delivers each packet to the nearest one. DNS root servers and CDNs use it, and every subnet has a Subnet-Router anycast address: its prefix with an all-zero interface ID.
Global addresses are valid everywhere. Link-local addresses are valid only on one link, which is why they need a zone ID. Multicast addresses carry an explicit 4-bit scope field.
Map of the whole address space
Who receives the packet?
Multicast address builder
Special-purpose prefixes
Select a row to open its example in the explorerSubnetting without the arithmetic pain
A site usually gets a /48 or a /56, and every LAN gets a /64. Subnetting is mostly counting in hex.
Registries allocate large blocks to ISPs, commonly a /32 or bigger. ISPs assign each customer a prefix: often a /48 for a business and a /56 for a home, sometimes only a /64, which allows just one LAN. Each LAN uses a /64 so that SLAAC can work.
The bits between the site prefix and bit 64 are the subnet ID. A /48 leaves 16 of them, so 65,536 LANs. A /56 leaves 8, so 256 LANs. Stick to multiples of 4 bits (nibble boundaries): then every subnet ID is whole hex digits, and reverse DNS delegation lines up with ip6.arpa labels.
You never squeeze hosts into a subnet. Router-to-router links often use /127 (RFC 6164) and loopbacks /128, but a LAN is always a /64.
Sandbox · Subnet planner
| # | Subnet | First address | Last address |
|---|
How the sizes nest
Interface IDs, EUI-64 and privacy
The last 64 bits identify the interface. How a device picks them decides whether websites can follow it from network to network.
The original method, modified EUI-64, builds the interface ID from the MAC address: split the 48-bit MAC in half, insert ff:fe, and flip the universal/local bit. It's deterministic, so a laptop would carry the same interface ID into every network it joins, which is a ready-made tracking handle.
Current systems use stable, opaque IDs (RFC 7217, recommended by RFC 8064): a hash of the prefix, the interface and a device secret. The result stays the same on one network and differs on every other. For outgoing connections they also add temporary addresses (RFC 8981) with random IDs that are replaced regularly, typically daily.
Every unicast address also gets a solicited-node multicast group, ff02::1:ffXX:XXXX, built from its last 24 bits. Neighbor Discovery sends its questions there, so almost no other host has to look at them.
Sandbox · Interface ID generator
The IPv6 header
Forty bytes and eight fields. No checksum, no options, no fragmentation fields: anything extra moves into extension headers.
The base header is always 40 bytes, so routers can parse it at a fixed offset. Compared with IPv4, four things went away: the header length (it never changes), the header checksum (links and TCP/UDP already check integrity, so routers no longer recompute it at every hop), the fragmentation fields (moved to the Fragment extension header) and options (moved to extension headers).
Several fields were renamed: TTL is now Hop Limit, Protocol is Next Header, Type of Service is Traffic Class, and Total Length became Payload Length, which no longer counts the base header. One field is new: the 20-bit Flow Label lets a source tag packets of one flow so routers can keep them on one path without parsing deeper.
Routers never fragment IPv6 packets. If a packet is too big for the next link, the router drops it and sends ICMPv6 Packet Too Big to the source, and every link must carry at least 1,280 bytes.
Sandbox · Header builder
Hover a byte to see which field it belongs to. Bytes 0–3 mix fields that don't align to byte boundaries.
What changed from IPv4
Sandbox · Extension header chain
Each header's Next Header field names what comes after it, forming a chain that ends at TCP, UDP or ICMPv6. Add headers and see whether the order follows RFC 8200.
| Header | NH value | Who processes it |
|---|---|---|
| Hop-by-Hop Options | 0 | Potentially every router on the path (many only do so when configured, and some drop such packets). Must come first. |
| Destination Options | 60 | Before a Routing header: each listed waypoint. At the end: only the final destination. |
| Routing | 43 | Each node listed in it (segment routing uses type 4; type 0 is deprecated). |
| Fragment | 44 | Only the destination, which reassembles. |
| Authentication (AH) | 51 | The destination (IPsec). |
| Encapsulating Security Payload (ESP) | 50 | The destination; everything after it is encrypted. |
ICMPv6 and Neighbor Discovery
ICMPv6 does far more than ping. Neighbor Discovery replaces ARP, finds routers, and checks that nobody else already uses an address.
ICMPv6 (Next Header 58) carries error messages (types 1–127) and informational messages (128–255). IPv6 cannot work without it, so it must never be blocked wholesale.
Neighbor Discovery (RFC 4861) uses five ICMPv6 messages: Router Solicitation (133), Router Advertisement (134), Neighbor Solicitation (135), Neighbor Advertisement (136) and Redirect (137). They're sent with Hop Limit 255, and receivers drop any that arrive with less, which proves the sender is on the same link.
- Address resolution: to learn a neighbor's MAC, a host sends a Neighbor Solicitation to the neighbor's solicited-node group, and the target replies with a Neighbor Advertisement.
- Neighbor Unreachability Detection keeps checking that neighbors are still there; cache entries move through REACHABLE, STALE, DELAY and PROBE.
- Duplicate Address Detection: before using any address, a host asks for it from the unspecified address
::. If anyone answers, the address is taken.
Simulation · One LAN, frame by frame
| No. | Time | Source | Destination | Protocol | Info |
|---|
ICMPv6 message types you'll meet
| Type | Message | Used for |
|---|---|---|
| 1 | Destination Unreachable | No route, address unreachable, port unreachable, administratively prohibited |
| 2 | Packet Too Big | Path MTU Discovery. Must never be filtered. |
| 3 | Time Exceeded | Hop Limit reached zero (traceroute), or reassembly timed out |
| 4 | Parameter Problem | Bad header field, unknown Next Header or option |
| 128 / 129 | Echo Request / Reply | ping |
| 130 / 131 / 132 / 143 | Multicast Listener Query / Report / Done / MLDv2 Report | Joining multicast groups; switches snoop them |
| 133 / 134 | Router Solicitation / Advertisement | Finding routers, prefixes, MTU and DNS (SLAAC) |
| 135 / 136 | Neighbor Solicitation / Advertisement | Address resolution, unreachability detection, DAD |
| 137 | Redirect | A router pointing a host to a better first hop |
Getting an address: SLAAC and DHCPv6
Routers advertise prefixes and hosts build their own addresses. DHCPv6 is optional, and the Router Advertisement decides who uses it.
SLAAC (stateless address autoconfiguration, RFC 4862): the router's advertisement carries a /64 prefix with the A flag set, and each host appends its own interface ID. No server keeps a record. DNS servers can travel in the advertisement itself (the RDNSS option, RFC 8106).
DHCPv6 (RFC 8415) can hand out addresses like DHCP does for IPv4, and identifies clients by a DUID rather than a MAC. It never provides a default gateway: that always comes from Router Advertisements. Two RA flags steer hosts: M (managed) says addresses are available from DHCPv6, and O (other) says other settings, such as DNS, are.
Android doesn't implement stateful DHCPv6 address assignment, so a network that turns SLAAC off leaves Android devices with no global address. Home routers get their LAN prefixes through DHCPv6 prefix delegation: the ISP delegates, say, a /56, and the router numbers each LAN as a /64 inside it.
Sandbox · Router Advertisement flags
DHCPv6 conversations
Routing and forwarding
Routers pick the most specific matching route, count the hop limit down, and never fragment.
- Longest prefix match: a destination can match many routes, and the one with the longest prefix wins.
::/0is the default route. - Link-local next hops are normal in IPv6. Routing protocols such as OSPFv3 and RIPng exchange routes over link-local addresses, so a routing table is full of
fe80::…next hops. - Hop Limit: every router subtracts one. At zero the packet is dropped and ICMPv6 Time Exceeded goes back to the source, which is exactly how traceroute works.
- Path MTU Discovery: when a packet is too big for a link, the router drops it and returns Packet Too Big with the link's MTU, and the source sends smaller packets. If a firewall blocks those messages, big transfers silently hang.
The routing protocols are the familiar ones, extended: OSPFv3 (RFC 5340, multicast ff02::5 and ff02::6), IS-IS, BGP with multiprotocol extensions, RIPng (UDP 521, ff02::9) and EIGRP for IPv6 (ff02::a).
Sandbox · Longest prefix match
| Route | Next hop | Interface | Bits matched |
|---|
Add your own route
Simulation · Hop limit and path MTU
Living with IPv4
Most networks run both protocols. Where they can't, tunnels and translators bridge the gap.
Dual stack is the default: devices run IPv4 and IPv6 side by side. Clients prefer IPv6 and fall back quickly with Happy Eyeballs (RFC 8305), so a broken IPv6 path costs a quarter of a second instead of a long timeout.
Tunnels carry one protocol inside the other: IPv6 in IPv4 (6in4, IP protocol 41; 6rd) or IPv4 in IPv6 (DS-Lite). Translation lets IPv6-only clients reach IPv4-only servers: DNS64 makes up AAAA records from A records using the prefix 64:ff9b::/96, and NAT64 translates the packets. 464XLAT adds a small translator on the device (CLAT), so IPv4-only apps keep working on IPv6-only mobile networks.
More networks now run IPv6-mostly: devices that cope without IPv4 switch it off when the DHCP server offers option 108 (RFC 8925).
Simulation · Happy Eyeballs race
Simulation · Tunnels and translators
Sandbox · IPv4-embedded IPv6 addresses (RFC 6052)
Reverse it: pull the IPv4 address out of an IPv6 address
Adoption today
On 28 March 2026, for the first time, more than half of the users reaching Google did so over IPv6.
“IPv6 adoption” depends on who is measuring. Google counts users who reach Google over IPv6. APNIC Labs measures how many users are IPv6-capable, using a worldwide ad campaign. Cloudflare counts the HTTP requests that arrive at its network over IPv6. Weekends score higher than weekdays, because home and mobile networks have better IPv6 than many offices.
Countries differ enormously. Large operators that switched early, often with a national push, put France and India far ahead, while other countries remain in single digits. The routing table keeps growing too: about 241,800 IPv6 prefixes in January 2026, against roughly 1.05 million for IPv4.
Share of users reaching Google over IPv6
Published milestones from Google's IPv6 statistics, as reported by the Internet Society, CircleID and The Register. Many are weekend peaks; weekday values run a few points lower.
Same month, different rulers
Global IPv6 share by measurement, April 2026. Google's figure is its first day above 50% (28 March).
Adoption varies by country
Internet Society Pulse, April 2026. The line marks Pulse's global average across its sources.
Routes in the global BGP table
January of each year, from Geoff Huston's annual BGP report. For scale: about 300 IPv6 prefixes in 2002 and 9,500 at World IPv6 Launch in 2012 (RIPE Labs).
Security
IPv6 is neither more nor less secure than IPv4. It is different: vastly more addresses, more reliance on ICMP, and local-link features that can be spoofed.
- No NAT doesn't mean no firewall. A stateful firewall that drops unsolicited inbound traffic gives the protection people credited to home NAT, and you can still open specific services on purpose.
- Don't block all ICMPv6. RFC 4890 lists what must pass: at least Packet Too Big, Destination Unreachable, Time Exceeded and Parameter Problem, and Neighbor Discovery must work on the link.
- Rogue Router Advertisements are the classic LAN problem, often by accident when a device starts acting as a router. Switches can enforce RA Guard (RFC 6105) and DHCPv6 Guard.
- Scanning a /64 blindly is hopeless, but attackers guess patterns instead:
::1, low numbers, EUI-64 IDs from known vendors, names from DNS. Don't treat an obscure address as access control.
Sandbox · What if you block this ICMPv6 message?
How long does a blind scan take?
Where attackers look instead
Real IPv6 reconnaissance narrows the space before sending a single probe:
- Low-byte addresses such as
::1,::10and::53that admins type by hand. - EUI-64 IDs: the
ff:fepattern plus a vendor's MAC prefix leaves only 24 unknown bits. - Embedded IPv4 addresses like
::192.0.2.10and service ports written as hex. - DNS zones, reverse
ip6.arpawalking, certificate transparency logs and server logs.
Defense: random or RFC 7217 IDs for hosts, deliberate naming for servers, and firewalls that don't depend on secrecy.
Threats and fixes
Practice terminal
Real commands on a simulated network. Try Linux, Windows and a Cisco router side by side.
Tasks
Click a command to paste it. ↑ and ↓ recall history; type help for everything this shell knows.
Check yourself
Twenty questions covering everything above. Your progress stays in this browser.
IPv6 quiz
Reference
The numbers and names you'll look up again.
Well-known multicast groups
| Address | Group |
|---|---|
| ff02::1 | All nodes on the link |
| ff02::2 | All routers on the link |
| ff02::5 / ff02::6 | OSPFv3 all SPF routers / designated routers |
| ff02::9 | RIPng routers |
| ff02::a | EIGRP routers |
| ff02::d | PIM routers |
| ff02::16 | MLDv2-capable routers (where MLDv2 reports go) |
| ff02::fb | Multicast DNS (mDNS) |
| ff02::1:2 | All DHCPv6 relay agents and servers |
| ff02::1:3 | LLMNR |
| ff05::1:3 | All DHCPv6 servers (site scope) |
| ff0x::101 | NTP, in any scope x |
| ff02::1:ff00:0/104 | Solicited-node groups: one per last-24-bits value |
Next Header values
| Value | Protocol or header |
|---|---|
| 0 | Hop-by-Hop Options |
| 4 | IPv4 (encapsulated, e.g. DS-Lite) |
| 6 | TCP |
| 17 | UDP |
| 41 | IPv6 (encapsulated, e.g. 6in4) |
| 43 | Routing |
| 44 | Fragment |
| 50 / 51 | ESP / Authentication Header |
| 58 | ICMPv6 |
| 59 | No Next Header |
| 60 | Destination Options |
| 89 | OSPF (OSPFv3) |
RFCs worth knowing
| RFC | Topic |
|---|---|
| 8200 | IPv6 specification (Internet Standard 86) |
| 4291 | Addressing architecture |
| 5952 | Canonical text representation |
| 4443 | ICMPv6 |
| 4861 / 4862 | Neighbor Discovery / SLAAC |
| 8415 | DHCPv6 |
| 8106 | DNS options in Router Advertisements |
| 7217 / 8981 | Stable opaque and temporary interface IDs |
| 4193 | Unique local addresses |
| 6052 / 6146 / 6147 | IPv4-embedded addresses, NAT64, DNS64 |
| 6877 / 6333 | 464XLAT / DS-Lite |
| 8305 | Happy Eyeballs v2 |
| 8201 | Path MTU Discovery for IPv6 |
| 4890 | ICMPv6 filtering recommendations |
| 6164 | /127 on point-to-point links |
| 3849 / 9637 | Documentation prefixes |
Glossary
No terms match that filter.