IPv4 uses 32-bit addresses, so it can name about 4.3 billion of them. IPv6 uses 128-bit addresses instead, written in hexadecimal with colons. That bigger space is the reason IPv6 exists, because the global free pool of IPv4 addresses ran out on 3 February 2011. The header changed too. IPv6 uses a fixed 40-byte header with no checksum, and routers never fragment its packets. It also drops broadcast entirely and replaces ARP with Neighbor Discovery. One popular claim is wrong, though. IPsec is not mandatory in IPv6 any more; the current node requirements make it a SHOULD, not a MUST.
IPv4 and IPv6 are two versions of the Internet Protocol, the layer that moves packets between networks. Most comparisons stop at “IPv6 has longer addresses”. That is true, but it hides the parts that change how a network actually behaves.
This guide checks every claim against the RFCs that define the two protocols. That matters here, because several popular IPv4 vs IPv6 tables repeat errors. The most common one says IPv6 has mandatory built-in security. It did once, but the standard changed in 2011.
The layer itself sits in the network layer of both reference models. Our OSI model vs TCP/IP model guide shows where that layer fits.
Why IPv6 Had to Exist
An IPv4 address is 32 bits long. So the whole address space holds 2^32 values, which is 4,294,967,296 addresses. That sounded enormous in 1981, when RFC 791 defined the protocol.
It was not enough, though. Because phones, laptops, servers and sensors all needed addresses, the pool shrank year after year. The regional registries then handed out the remaining blocks.
The moment arrived on 3 February 2011. That day the Number Resource Organization announced that “the free pool of available IPv4 addresses is now fully depleted”. In other words, no central reserve was left to hand out.
IPv6 answers that shortage directly. Its addresses are 128 bits long, so the space holds 2^128 values, roughly 3.4 × 10^38. The current IPv6 specification, RFC 8200, dates from July 2017.
How IPv4 Addressing Works
IPv4 writes an address as four decimal numbers separated by dots. Each number is one octet, so it runs from 0 to 255. This format is called dotted-decimal notation.
A documentation example is 192.0.2.1. RFC 5737 reserves the 192.0.2.0/24 range for examples like this one, so it never routes on the real internet. Using it here avoids pointing readers at somebody’s live machine.
Early IPv4 split addresses into fixed classes. That scheme wasted space, so the internet moved to classless addressing with variable-length masks. Our classful vs classless addressing guide covers that shift in detail.
Address scarcity also shaped IPv4 networks in practice. As a result, most homes and offices sit behind Network Address Translation, or NAT. Many private devices then share one public address, which stretches the pool further.
How IPv6 Addressing Works
IPv6 writes an address as eight groups of four hexadecimal digits. Colons separate the groups, and each group carries 16 bits. Eight groups of 16 bits give the full 128.
The documentation range here is 2001:db8::/32, reserved by RFC 3849. A typical example is 2001:db8::1. The double colon stands for a run of zero groups, a shorthand the next section explains.
Most IPv6 unicast addresses split into a network prefix and an interface identifier. RFC 4291 requires that identifier to be 64 bits long for most unicast addresses. So a standard subnet uses a /64 prefix, and each host fills in the remaining 64 bits.
That layout supports automatic configuration. A host can build its own address from the router’s advertised prefix plus its own identifier. RFC 4862 calls this stateless address autoconfiguration, or SLAAC.
IPv4 vs IPv6: Comparison Table
The table below lines up the two protocols field by field. Every value traces back to the RFC named in the text, rather than to secondary summaries.
| Aspect | IPv4 | IPv6 |
|---|---|---|
| Address length | 32 bits | 128 bits |
| Address space | 2^32, about 4.3 billion | 2^128, about 3.4 × 10^38 |
| Notation | Dotted decimal, four octets | Hexadecimal, eight 16-bit groups separated by colons |
| Documentation example | 192.0.2.1 (RFC 5737) | 2001:db8::1 (RFC 3849) |
| Header size | 20 to 60 bytes, variable | 40 bytes, fixed |
| Header fields | 12 fixed fields, plus options | 8 fields |
| Options | Carried inside the header | Moved to separate extension headers |
| Header checksum | Yes, recomputed at every hop | None; the UDP checksum becomes mandatory instead |
| Hop counter | Time to Live, 8 bits | Hop Limit, 8 bits |
| Fragmentation | Sender or routers along the path | Source node only, never routers |
| Minimum packet size | Hosts must accept 576 bytes | Every link must carry 1280 bytes |
| Broadcast | Supported | Removed; multicast covers its job |
| Address resolution | ARP | Neighbor Discovery (NDP) |
| Router discovery and redirect | Separate ICMP mechanisms | Folded into Neighbor Discovery |
| Address configuration | Manual or DHCP | SLAAC, DHCPv6, or manual |
| Subnetting | Classless, with variable-length masks | Prefix length; /64 is standard for a subnet |
| IPsec | Optional, added when needed | A SHOULD for all nodes; was a MUST before RFC 6434 |
| Quality of service fields | Type of Service byte | Traffic Class, plus a 20-bit Flow Label |
| NAT | Common, to conserve addresses | Not needed for address conservation |
| Defining standard | RFC 791, September 1981 | RFC 8200, July 2017 |
| Header processing | Variable length, so routers parse more | Fixed base header, simpler to parse |
Two rows trip up most readers. IPv6 has no broadcast, and it has no header checksum. Both are deliberate removals, and the next sections explain why.
What Changed in the Header
An IPv4 header carries a field called Internet Header Length. RFC 791 sets its minimum value at 5, counted in 32-bit words, which gives 20 bytes. The same RFC also states that “an internet header may be up to 60 octets” once options are added.
IPv6 took a different route. Its base header is always 40 bytes, and RFC 8200 explains the goal plainly: “Some IPv4 header fields have been dropped or made optional”. Optional data now lives in extension headers, chained after the base header.
The checksum is the clearest example. IPv4 recomputes a header checksum at every hop, because the Time to Live field changes each time. In contrast, IPv6 drops that checksum entirely.
The cost moves elsewhere, though. RFC 8200 states that, for IPv6, “the UDP checksum is not optional”. So the transport layer carries the integrity check instead. Our TCP vs UDP guide covers those two transports.
Fragmentation changed too. In IPv4, a router can split a packet that is too large for the next link. RFC 8200 rules that out: “fragmentation in IPv6 is performed only by source nodes, not by routers along a packet’s delivery path”.
That rule needs a guaranteed floor, otherwise a sender could never pick a safe size. RFC 8200 sets it: “IPv6 requires that every link in the Internet have an MTU of 1280 octets or greater”. By comparison, RFC 791 only requires IPv4 hosts to accept 576 octets.
The IPsec Myth
Many comparison tables claim IPv6 has mandatory, built-in IPsec. Our own earlier version of this page said so too. However, that claim is out of date.
IPsec is a suite that authenticates and encrypts IP traffic. Early IPv6 node requirements did make it mandatory. However, RFC 6434 downgraded it, and the current RFC 8504 keeps that change.
RFC 8504 describes IPsec support as “a SHOULD for all IPv6 nodes”. In standards language, SHOULD means recommended, not required. So an IPv6 node can comply without implementing IPsec at all.
The reverse claim is wrong as well. IPsec runs over IPv4 too, and IPv4 VPNs use it every day. Either way, neither protocol encrypts traffic by default.
So treat security as a separate decision on both versions. For encryption at the application layer instead, see our SSL vs TLS guide.
Reading an IPv6 Address
Long hexadecimal strings invite mistakes, so RFC 5952 fixes one canonical text form. It rules that hexadecimal digits “MUST be represented in lowercase”. As a result, logs and comparisons stay consistent.
Zero groups get compressed with a double colon. RFC 5952 adds that this shorthand “MUST be used to its maximum capability”. So the longest run of zero groups always collapses.
Take the address 2001:db8:0:0:0:0:0:1 as an example. Its six middle zero groups compress into one double colon. The canonical form then reads 2001:db8::1.
The double colon may appear only once in an address. Otherwise a reader could not tell how many zero groups each one hides. Leading zeros inside a group also drop, so 0db8 becomes db8.
What IPv6 Actually Improves
The address space is the headline gain, and it is real. Every device can hold a globally unique address again. As a result, NAT is no longer needed just to conserve addresses.
Configuration also gets simpler. For example, with SLAAC a host builds its own address without a DHCP server. Many networks still run DHCPv6, though, when they want central control.
Local discovery changed as well. RFC 4861 says Neighbor Discovery corresponds to a combination of three IPv4 mechanisms. Those are ARP, ICMP Router Discovery and ICMP Redirect. So one protocol now does the work of three.
Broadcast is gone too. RFC 4291 states that “There are no broadcast addresses in IPv6, their function being superseded by multicast addresses”. That cuts traffic every host on a link would otherwise have to process.
Speed is a different matter. IPv6 is not inherently faster than IPv4, despite frequent claims. Instead, real performance depends on the path, the hardware and the configuration.
Running Both Together
IPv4 and IPv6 cannot talk to each other directly. Their headers and addresses differ completely, so an IPv6-only host cannot reach an IPv4-only server alone.
Most networks therefore run dual stack. Each device holds both an IPv4 address and an IPv6 address, then uses whichever one the destination supports. This is the most common transition approach.
Two other tools fill the gaps. Tunnelling wraps IPv6 packets inside IPv4 packets, so they can cross IPv4-only links. In contrast, translation rewrites packets between the two protocols at a gateway.
A sensible migration starts small. To begin with, audit which devices and services already support IPv6. Then enable dual stack, and move services across gradually, rather than switching everything at once.
The physical and link details stay the same throughout. Hardware addresses still identify each interface on the local link. Our MAC address vs IP address guide explains how the two layers work together.
Interview Questions
Frequently Asked Questions
Wrapping Up
IPv4 vs IPv6 starts with address length, 32 bits against 128. That gap exists because the IPv4 free pool ran out in 2011.
However, the deeper changes sit in the header and on the local link. IPv6 uses a fixed 40-byte header with no checksum. Routers never fragment its packets, broadcast is gone, and Neighbor Discovery replaces ARP.
Finally, remember the correction. IPsec is a SHOULD in IPv6, not a MUST, and it runs on IPv4 too. Neither version is secure by default. For more networking comparisons, browse the CS Fundamentals hub.
simple and good stuff