The sentence “we don’t use IPv6” is almost never true on campus networks. Windows, macOS and Linux clients ship with IPv6 enabled by default and talk with a link-local address on every interface; IPv6 traffic is already flowing on your network, it’s just that nobody planned it. The cost of putting it off is usually paid one day with a fake Router Advertisement. Below we explain a dual-stack campus migration with actionable steps, from the address plan to the client addressing method and from RA protection to the phased rollout order.
IPv6 migration strategies for a campus
There are three basic approaches on a campus network:
| Approach | How does it work? | When is it suitable? |
|---|---|---|
| Dual-stack | Every segment carries both IPv4 and IPv6 | The default path on enterprise campuses |
| IPv6-only + NAT64/DNS64 | The client uses only IPv6 and reaches IPv4 destinations through translation | Newly built segments with a well-known application inventory |
| Tunneling | IPv6 packets are carried over IPv4 | A temporary solution; not recommended within a campus |
In enterprise networks, old printers, industrial devices and IPv4-dependent applications keep living for years. That’s why the practical goal isn’t to “turn off IPv4” but to operate IPv6 at the same quality and security level as IPv4. From here on, the dual-stack model is the basis. That’s what’s realistic.
Address plan: a sample table from block to VLAN
The logic of an IPv6 address plan is different from IPv4. There’s no shortage of addresses; what’s scarce is readability and the ability to aggregate routes under a single prefix. The first plan of teams thinking with the IPv4 habit of “small subnet, less waste” usually ends up in the bin. Basic rules:
- Every end-user and server segment is a /64. SLAAC works only with a /64 prefix; a smaller subnet creates client incompatibility.
- /127 can be used on point-to-point router links (RFC 6164); loopback addresses are /128.
- Subdivision is done on nibble (4-bit, single hexadecimal digit) boundaries; prefixes stay readable and DNS reverse zones are easy to define.
- Decide the prefix source from the start. A provider-assigned (PA) prefix requires renumbering when the provider changes. Provider-independent address space (via RIPE NCC) removes this risk but requires announcing it with BGP. Find out in advance what length of prefix your provider gives (for example /48 or /56).
The example below assumes a /48 has been allocated to a site from the 2001:db8::/32 block reserved for documentation (RFC 3849). The first digit of the fourth hextet shows the zone (0 infrastructure, 1 building A, 2 building B, f data center), and the last three digits the VLAN number:
| Use | Example prefix | Note |
|---|---|---|
| Loopbacks | 2001:db8:4a00:0::/64 | /128 per device |
| Point-to-point links | 2001:db8:4a00:1::/64 | /127 per link |
| Management (VLAN 100) | 2001:db8:4a00:100::/64 | Infrastructure zone |
| Building A users (VLAN 20) | 2001:db8:4a00:1020::/64 | Building A summary: /52 |
| Building A voice (VLAN 120) | 2001:db8:4a00:1120::/64 | Within the same /52 |
| Building B users (VLAN 20) | 2001:db8:4a00:2020::/64 | Building B summary: /52 |
| Servers (VLAN 50) | 2001:db8:4a00:f050::/64 | Data center zone |
In this structure, all of building A’s segments are aggregated under 2001:db8:4a00:1000::/52; per-building route summarization and firewall objects are written with a single prefix. Reflecting the VLAN number in the address works if your existing IPv4 VLAN layout is consistent. If the layout is messy, tidy it up first; an IPv6 plan built on scattered VLANs makes the mess permanent. For that, the VLAN design in enterprise networks article is the starting point.
SLAAC, stateless DHCPv6 and stateful DHCPv6
IPv6 clients can get an address in three ways. The overlooked detail: the default gateway comes via Router Advertisement (RA) in every case. DHCPv6 doesn’t carry a default route.
| Method | Address | DNS | Watch out for |
|---|---|---|---|
| SLAAC | The client generates it itself | RDNSS within the RA (RFC 8106) | Older clients lack RDNSS support |
| SLAAC + DHCPv6 stateless | The client generates it itself | DHCPv6 | The server doesn’t keep address records |
| DHCPv6 stateful | The DHCPv6 server assigns it | DHCPv6 | Android doesn’t take a stateful address (IA_NA) |
The common choice on enterprise Windows networks is stateful DHCPv6 on Windows computer segments (for address logging and dynamic DNS registration), and SLAAC on mobile device and guest segments. The reason is Android: it doesn’t support getting an address (IA_NA) via DHCPv6, and on a segment with only the M flag set and the A flag cleared, it can’t obtain an IPv6 address. Test the current behavior of mobile operating systems in the pilot phase. With stateful DHCPv6, reservations are made not by MAC address but by DUID; the DUID can change when the operating system is reinstalled. On SLAAC segments, clients generate temporary addresses for privacy. In an incident investigation, the question “which IP was which device” can only be answered with the ND table, NetFlow or NAC records. Collect these records from the start.
Router Advertisement, M/O flags and RA Guard
The RA message tells the client three things: the default gateway, prefix information (whether SLAAC is allowed, via the A flag) and whether DHCPv6 will be used. The M (Managed) flag indicates the address will be obtained from DHCPv6, and the O (Other) flag that only other information such as DNS will come from DHCPv6. An example SVI using stateful DHCPv6 on Cisco IOS XE:
ipv6 unicast-routing
!
interface Vlan20
description USERS-BUILDING-A
ipv6 address 2001:db8:4a00:1020::1/64
ipv6 nd managed-config-flag
ipv6 nd other-config-flag
ipv6 nd prefix 2001:db8:4a00:1020::/64 no-autoconfig
ipv6 dhcp relay destination 2001:db8:4a00:f050::11
ipv6 dhcp relay destination 2001:db8:4a00:f050::12
The no-autoconfig parameter clears the A flag on the prefix; the client doesn’t generate a SLAAC address. DHCPv6 relay logic is similar to ip helper-address in IPv4. For relay and rogue server protection on the IPv4 side, see the DHCP relay and DHCP snooping article.
In IPv6, a more common risk than a rogue DHCP server is the rogue RA. If a computer with internet connection sharing enabled or a misconfigured virtual machine sends RAs, clients may choose it as their default gateway. This risk exists even on networks that haven’t yet deployed IPv6, because IPv6 is enabled on clients. It’s one of the possible causes of the complaint “some users connect slowly to some sites.” Protection is provided on access ports with RA Guard (RFC 6105) and DHCPv6 Guard:
ipv6 nd raguard policy RA-HOST
device-role host
!
ipv6 dhcp guard policy DHCP6-CLIENT
device-role client
!
interface range GigabitEthernet1/0/1 - 46
ipv6 nd raguard attach-policy RA-HOST
ipv6 dhcp guard attach-policy DHCP6-CLIENT
Some implementations of RA Guard may not catch RA messages fragmented with IPv6 extension headers (the topic is covered in RFC 7113). Verify how the access switch’s software version behaves against these evasions from the vendor’s documentation.
The importance of ICMPv6 and Neighbor Discovery
Blocking ICMP wholesale in IPv4 is a bad habit, but the network keeps working. In IPv6, if ICMPv6 is blocked, the network doesn’t work. Neighbor Discovery (ND) replaces ARP and runs on ICMPv6 messages: Router Solicitation (133), Router Advertisement (134), Neighbor Solicitation (135), Neighbor Advertisement (136) and Redirect (137). In addition, routers don’t fragment packets in IPv6; for a packet larger than the MTU along the path, a “Packet Too Big” (type 2) message returns to the source. If this message is blocked on the firewall, small requests work and large transfers hang. It’s one of the hardest faults to diagnose.
The reference for ICMPv6 in firewall and ACL design is RFC 4890. In outline: Destination Unreachable (1), Packet Too Big (2), Time Exceeded (3) and Parameter Problem (4) should be allowed in transit; ND messages must always work on the local link. ND uses solicited-node multicast addresses for address resolution, and clients join these groups with MLD (ICMPv6 types 130–132 and 143). Don’t block MLD messages on the local link either; if MLD snooping is used on switches, test that ND isn’t affected.
Phased dual-stack migration order
Order matters. You put protection in place before routing, and policy before rolling out broadly:
- Inventory: List the IPv6 support, software versions and license requirements of switches, routers, firewalls, wireless controllers, load balancers and monitoring tools. Check whether IPAM and SIEM handle IPv6 addresses.
- Access protection: Apply RA Guard and DHCPv6 Guard to all access switches before IPv6 routing is enabled.
- Address plan and prefix: Get the plan table approved, and prepare DNS reverse zones.
- Backbone: Loopbacks, point-to-point links and dynamic routing (OSPFv3 or the IPv6 address family) are enabled at the core and distribution layers. For the campus routing structure, the OSPF design in a campus network article can be used as a reference.
- Security policy parity: An IPv6 counterpart is written on the firewall for every IPv4 rule. A firewall that is strict for IPv4 and open for IPv6 effectively eliminates segmentation. That’s how a back door opens.
- Services: DHCPv6 scopes, the IPv6 addresses of DNS servers and internal DNS records are prepared. For separating internal and external DNS, see the enterprise DNS architecture article.
- Pilot: Start with the IT team’s VLAN; application access, VPN, printing and monitoring are tested.
- Rollout: User VLANs are enabled building by building. Server segments and internal AAAA records are published after verifying that the service is really reachable over IPv6.
Verification commands and common mistakes
Every time a segment is enabled, do the following checks from the console:
! Cisco IOS XE
show ipv6 interface brief
show ipv6 interface Vlan20
show ipv6 neighbors
show ipv6 route
show ipv6 nd raguard policy RA-HOST
# Windows (PowerShell)
Get-NetIPAddress -AddressFamily IPv6
Get-NetRoute -AddressFamily IPv6 -DestinationPrefix ::/0
ping -6 2001:db8:4a00:f050::11
# Linux
ip -6 addr show
ip -6 route show default
rdisc6 eth0
The show ipv6 interface Vlan20 output shows the RA interval and the state of the M/O flags. rdisc6 on Linux (the ndisc6 package) lists RA messages on the segment; if you see RAs from an unexpected source, RA Guard is missing or has been bypassed.
Mistakes you may encounter:
- Using a longer prefix instead of /64 on user segments and breaking SLAAC.
- Blocking ICMPv6 wholesale out of IPv4 habit.
- Leaving IPv6 rules incomplete on the firewall or running them with a default permit.
- Publishing an AAAA record before the service is ready over IPv6; clients trying IPv6 first and falling back to IPv4 with a delay.
- Planning DHCPv6 reservations by MAC address.
- Not accounting for IPv6 address–device mapping in log and incident investigation processes.
Designing the address plan together with the existing VLAN and routing structure can be handled within VLAN, routing and network segmentation work, and producing the migration roadmap within enterprise network consulting.
Frequently asked questions
If we don’t use IPv6, do we need RA Guard?
Yes. IPv6 is enabled by default on Windows, macOS and Linux clients. If a device on the network sends a fake Router Advertisement, clients can use it as their IPv6 default gateway. Apply RA Guard to access ports before enabling IPv6 routing.
Why do we have to use /64 on user VLANs?
SLAAC generates a 64-bit interface identifier and works only with /64 prefixes. If a longer prefix is used, SLAAC is disabled and some clients can’t get an address. Because the IPv6 address space is vast, using /64 isn’t considered waste.
Should I choose SLAAC or DHCPv6?
If you want address logging and dynamic DNS updates, stateful DHCPv6 is suitable for Windows computer segments. Because Android devices don’t get an address (IA_NA) via DHCPv6, use SLAAC on mobile and guest segments. In both cases, the default gateway comes via Router Advertisement.
Which ICMPv6 messages should be allowed on the firewall?
According to RFC 4890, Destination Unreachable, Packet Too Big, Time Exceeded and Parameter Problem messages should be allowed in transit. Neighbor Discovery messages (133–137) must always work on the local link. If Packet Too Big is blocked, large packet transfers hang.