IS-IS for IPv6
Explain how IS-IS carries IPv6 with the TLVs from RFC 5308. Explain why single-topology IS-IS breaks when IPv4 and IPv6 topologies differ, and how multi-topology (RFC 5120) fixes it. Build NETs and configure IS-IS for IPv6 with multi-topology on FRR. Read the IS-IS database and routing table for both address families.
Take the quiz · pass mark 70% · 9 questions
Explanation
IS-IS is the other link-state IGP, and many service providers prefer it to OSPF. It runs directly over the link layer (not over IP), so it was never tied to an IP version: RFC 1195 added IPv4 to the original OSI protocol, and adding IPv6 needed only new TLVs. That makes it a natural fit for dual-stack and IPv6-only cores.
IS-IS basics in one page
- Routers are Intermediate Systems. Each has a NET (Network Entity Title) such as
49.0001.1920.0000.2021.00: area49.0001(49 = private AFI), a 6-byte System ID, and selector00. - Two levels: Level 1 (inside an area) and Level 2 (the backbone between areas). Small and medium networks often run a single area at Level 2 only.
- Each router floods one LSP (link-state PDU) per level, made of TLVs (type-length-value). Adding a protocol means adding TLVs.
- Use wide metrics (
metric-style wide, RFC 5305): 24-bit link metrics instead of the old 6-bit ones. A link advertised with the maximum value 2^24 − 1 = 16,777,215 is excluded from normal SPF, so the largest usable link metric is 16,777,214.
A common way to build the System ID is from the router's IPv4 loopback with each octet padded to three digits: 192.0.2.21 → 192000002021 → 1920.0000.2021. On an IPv6-only network you can use any unique 12 hex digits instead, but a documented scheme avoids duplicates.
IPv6 TLVs (RFC 5308)
| TLV | Name | Purpose |
|---|---|---|
| 129 | Protocols Supported | NLPID 0x8E announces IPv6 (0xCC is IPv4) |
| 232 | IPv6 Interface Address | Link-local in Hellos, non-link-local in LSPs |
| 236 | IPv6 Reachability | Prefixes with a 32-bit metric, an Up/Down bit and an External bit |
In Hellos, TLV 232 carries the link-local address, which becomes the next hop for IPv6 routes, just as in OSPFv3. Prefix metrics above 0xFE000000 (MAX_V6_PATH_METRIC) are not used in SPF.
Single topology and its problem
With plain RFC 5308, IS-IS runs one SPF for both IPv4 and IPv6: the shortest path tree is the same and both address families follow it. That only works if every link carries both protocols. Suppose one core link is IPv6-only (no IPv4 addresses). The SPF still uses it, so IPv4 traffic is routed onto a link that has no IPv4 next hop. Depending on the implementation, the adjacency is refused (losing the link for IPv6 too) or IPv4 routes through it are missing or black-holed. Either way, single topology forces identical topologies.
Multi-topology (RFC 5120)
Multi-topology IS-IS runs a separate SPF per topology:
| MT ID | Topology |
|---|---|
| 0 | Standard topology (IPv4 unicast) |
| 2 | IPv6 unicast |
New TLVs carry the topology: 229 (MT membership in Hellos and LSPs), 222 (MT IS reachability, the per-topology version of TLV 22), and 237 (MT IPv6 reachability, the per-topology version of TLV 236). Each link joins only the topologies it supports, so an IPv6-only link is used by MT 2 and ignored by MT 0.
Plan multi-topology from the start: changing from single to multi-topology on a live network needs care, because routers in different modes calculate different trees. Some implementations offer a transition mode.
Recent FRR releases add another approach: isis ipv4-over-ipv6-nexthop routes IPv4 over an interface with no IPv4 address using the neighbour's IPv6 link-local next hop (the same idea as BGP with RFC 8950, L4.04). It is new and implementation-specific; multi-topology is the standard answer.
Configuration (FRR isisd)
interface lo
ipv6 router isis CORE
isis passive
!
interface eth1
ip router isis CORE
ipv6 router isis CORE
isis network point-to-point
isis metric 10
!
interface eth2
ipv6 router isis CORE
isis network point-to-point
isis metric 10
!
router isis CORE
net 49.0001.1920.0000.2021.00
is-type level-2-only
metric-style wide
topology ipv6-unicast
eth2 is IPv6-only. topology ipv6-unicast turns on multi-topology for IPv6. Without it, FRR uses single topology. Useful commands: show isis neighbor, show isis database detail, show isis topology, show isis route.
On MikroTik RouterOS v7, IS-IS support has been added in recent releases and its feature set differs; check the manual for your version. VyOS uses FRR underneath (set protocols isis ...).
ictNetworks scenario
ictNetworks' new Cape Town PoP is built IPv6-only on the inside: its two core links carry only IPv6, and IPv4 is delivered to customers as a service (Level 6). The PoP runs IS-IS Level 2 with multi-topology. Router loopbacks follow the existing plan (3fff:a00::21/128 for cpt-cr1), and System IDs come from the documented 192.0.2.x router IDs used by OSPFv3 and BGP.
Drawing diagram…
Worked examples
Example 1: Building a NET
Router r2 has router ID 192.0.2.22. Padding gives 192.000.002.022, so the System ID is 1920.0000.2022 and with area 49.0001 the NET is 49.0001.1920.0000.2022.00.
Example 2: Two topologies
In the diagram, r1–r2 and r1–r3 are dual-stack and r2–r3 is IPv6-only. With multi-topology:
- MT 2 (IPv6) can use r1–r2–r3 at cost 10 + 10 = 20, which is less than the direct link (30), so IPv6 from r1 to r3 goes via r2.
- MT 0 (IPv4) cannot use r2–r3, so IPv4 from r1 to r3 uses the direct link at cost 30.
With single topology, both families would follow r1–r2–r3, and IPv4 would fail on the r2–r3 link.
Example 3: Metric limits
The largest wide link metric that is still used in SPF is 2^24 − 2 = 16,777,214. Advertising 16,777,215 removes the link from SPF (useful for traffic-engineering-only links).
Lab
Goal: Show that single-topology IS-IS fails with an IPv6-only link, then fix it with multi-topology.
You need: containerlab with three FRR nodes in the triangle from the diagram. Loopbacks 2001:db8:1000::21/128 to ::23/128 and 192.0.2.21/32 to .23/32; dual-stack links use 2001:db8:1000:1LLL::/127 and IPv4 /31s 192.0.2.100/31 (r1–r2) and 192.0.2.102/31 (r1–r3); r2–r3 has IPv6 only.
- Compute the three NETs and configure IS-IS Level 2 with wide metrics and no
topologycommand (single topology). Set the metrics from the diagram. - Show the adjacencies and LSPs (
show isis database detail) and identify TLVs 129 (with NLPID 0x8E), 232 and 236. Then show what happens to IPv4 from r1 to r3's loopback: route table, traceroute and your explanation. Record whether your FRR version refuses the adjacency or installs unusable IPv4 routes. - Add
topology ipv6-unicaston all three routers. Show TLVs 229, 222 and 237, and showshow isis topologyfor both topologies. - Verify that IPv6 from r1 to r3 goes via r2 (cost 20) while IPv4 uses the direct link (cost 30), with traceroutes for both.
- Set the r1–r3 metric to 16777215, show that the link disappears from SPF, and restore it.
What to submit: configurations, the NET calculation, database excerpts with TLVs marked, the single-topology failure evidence and the multi-topology traceroutes.
Summary
- IS-IS runs over the link layer and carries IPv6 with TLVs 232 and 236 and NLPID 0x8E (RFC 5308).
- Next hops for IPv6 routes are link-local, as in OSPFv3.
- Single topology uses one SPF for both families and requires every link to carry both.
- Multi-topology (RFC 5120) gives IPv6 its own SPF (MT ID 2, TLVs 229, 222, 237), so IPv6-only links work.
- Use wide metrics; the maximum value 16,777,215 removes a link from SPF.
- Next lesson (L4.04): BGP for IPv6.
Quiz imported · 9 questions · pass mark 70%. Lab questions are not auto-marked.