OSPF network types in detail

A /24 loopback that shows up as /32

The OSPF network type is an interface setting that controls how OSPF uses a link. It affects neighbour discovery, default timers, election roles and the connection records shared with other routers. It does not change the physical cable, link speed or IP subnet mask.

You configure Loopback1 on R3 with 172.16.3.1/24 to stand in for a test LAN and add it to OSPF. On R1, show ip route lists O 172.16.3.1/32, not 172.16.3.0/24. An extra network statement doesn't help. Later, one end of a router link is changed to point-to-point to remove DR election and the other end is left alone: the neighbours still reach Full, yet routes through the link disappear. Both puzzles are about the OSPF network type.

The DR/BDR lesson introduced broadcast and point-to-point. This lesson covers all the types side by side.

Six network types and their defaults

Start with one Ethernet connection: R2 Gi0/1 is 10.0.23.1/30 and R3 Gi0/0 is 10.0.23.2/30. Both interfaces belong to area 0. Their router IDs, 2.2.2.2 and 3.3.3.3, identify the routers in OSPF; they are separate from the port addresses.

Broadcast is the Ethernet default, even with only two attached routers. OSPF elects a designated router (DR) to organise database exchanges and a backup designated router (BDR) ready to replace it. The election does not make either router the required next hop for user traffic.

Point-to-point treats the connection as a link to one other router. The two routers synchronise directly without DR or BDR roles. That is useful on an Ethernet link connecting exactly two routers, but it must be a deliberate setting at both ends.

A Hello discovers and maintains neighbours. The Dead interval is the allowed time without a valid Hello before declaring the neighbour lost. Multicast sends a packet to a group, such as OSPF routers at 224.0.0.5. Unicast sends it to one specified neighbour address. The table shows Cisco IOS defaults; configured timer overrides can differ.

The types differ in three ways that matter in practice:

• DR election. Broadcast and non-broadcast elect a DR and BDR, wait for the dead interval (the wait timer: 40 s on broadcast, 120 s on non-broadcast) when an eligible interface comes up, unless an election event ends the wait sooner. A DR with a Full neighbour originates a type 2 LSA for the shared transit segment. The point-to-point types don't.
• Timers. The timers are part of the hello check, so two types with different default timers can't become neighbours unless you change the timers by hand.
• Neighbour discovery. On networks without multicast (Frame Relay, some cloud networks) the router can't find neighbours by itself, so you list them with the neighbor command.

Why shared addressing does not prove a direct path

A wide area network (WAN) connects sites. A hub-and-spoke transport connects each branch, or spoke, to a central hub. The spokes may share one IP subnet without having a direct connection to each other. Before selecting an OSPF network type, check which router pairs can communicate and whether the transport delivers the required multicast packets. An interface name such as Tunnel0 does not answer those questions by itself.

In this separate example, R1 is the hub. R2 and R3 are spokes on 10.0.50.0/24. The WAN is already configured to deliver multicast between R1 and each spoke. All three OSPF interfaces use point-to-multipoint, area 0 and matching 30/120 timers. The lines below show the resulting neighbour paths over that transport.

Point-to-multipoint models the reachable neighbour pairs as individual links and avoids a DR election. R1 becomes Full with both spokes. R2 does not need to become adjacent to R3 over a transport path that does not exist. OSPF creates host routes for the multipoint interface addresses; on R2, a route to 10.0.50.3/32 through 10.0.50.1 directs traffic towards R3 through the hub, despite the shared /24 addressing.

The trade-off is additional host routes to inspect. If multicast discovery is unavailable, Cisco's point-to-multipoint non-broadcast option needs configured neighbours. Non-broadcast multi-access (NBMA) is another option, but it retains DR/BDR roles and requires a suitable reachability design. Neither setting creates missing transport connectivity.

Why one link behaviour couldn't fit 1990s WANs

OSPF was designed for every kind of link in use in the early 1990s: Ethernet LANs where everyone hears everyone, leased lines with exactly two routers, and WAN clouds such as Frame Relay where routers share a subnet but can't multicast and may not all reach each other. One behaviour can't fit them all. On Ethernet a DR saves work; between two routers it is pointless; on a partial-mesh WAN a DR that can't reach every spoke would break flooding, which is what point-to-multipoint avoids.

Ethernet is also common between just two routers. Setting these links to point-to-point removes the DR election, its wait procedure and the type 2 LSA for the segment. The trade-off is that this setting assumes a two-router connection; do not apply it blindly to a shared LAN with several OSPF routers.

The loopback /32 surprise

A loopback is an interface created in software. Here, R3 owns address 172.16.3.1 with mask 255.255.255.0 (/24) on Loopback1. The diagram compares two advertisements from that same interface, not two physical cables. A host route (/32) matches one address; a subnet route (/24) matches the whole prefix.

A loopback has the network type LOOPBACK. OSPF treats it as a single host and advertises it as a /32 host route, whatever mask is configured. That is usually fine for router IDs and management addresses. When a loopback stands in for a whole subnet, as in labs, change its network type:

R3#show ip ospf interface Loopback1
Loopback1 is up, line protocol is up
  Internet Address 172.16.3.1/24, Area 0, Attached via Network Statement
  Process ID 1, Router ID 3.3.3.3, Network Type LOOPBACK, Cost: 1
  Loopback interface is treated as a stub Host

Setting the network type on Cisco IOS

Mismatches: which ones break the adjacency

The first row is the dangerous one, because show ip ospf neighbor looks healthy. Here is what happens on the R2–R3 link when R2's Gi0/1 is point-to-point and R3's Gi0/0 is still broadcast:

A link-state advertisement (LSA) describes part of the network. Routers carry these descriptions inside Link State Update (LSU) packets. The shortest path first (SPF) calculation then joins the descriptions into usable paths. Reaching Full means the neighbours have synchronised their databases; it does not guarantee that their descriptions of this connection agree. In the walkthrough below, R2 describes a direct router-to-router link while R3 describes a shared network. The missing matching connection prevents SPF from using that link.

R3#show ip ospf interface GigabitEthernet0/0 | include Network Type
  Process ID 1, Router ID 3.3.3.3, Network Type BROADCAST, Cost: 10

In practice this mismatch tends to appear after a change window: the point-to-point command went onto one router of the pair, and the change on the other router was skipped or rolled back.

Short lab: change the same link at both ends

In an isolated Cisco IOS or IOS XE lab, start R2 Gi0/1 and R3 Gi0/0 with broadcast OSPF, compatible addressing and working routes. Record the network type on each end. In R2's neighbour output, FULL/DR means the peer has the DR role; it does not describe R2's own role.

Enter ip ospf network point-to-point under R2 Gi0/1 and R3 Gi0/0. The change can interrupt the adjacency, so perform it in the lab rather than on a link carrying production traffic. Recheck both ends after convergence. They should show point-to-point, matching 10/40 timers if unmodified, and Full with a dash instead of a DR/BDR role.

Finally check a known remote LAN route and test the assigned host address. Full is a database state, not an end-to-end traffic test. If you restore broadcast, restore it on both ends with default ip ospf network and verify the new election result; the elected roles need not be the same as before.

Network type mistakes that hide behind a Full neighbour

• Changing the network type on one end of a link only.
• Expecting a DR on a point-to-point or point-to-multipoint network.
• Assuming a loopback is advertised with its configured mask.
• Forgetting the neighbor commands on a non-broadcast network, so no neighbours ever form.
• Thinking Ethernet links default to point-to-point because only two routers are connected. The default comes from the media (Ethernet = broadcast), not from how many routers are attached.

Exam tip: know the defaults (Ethernet = broadcast with DR, serial HDLC/PPP = point-to-point without DR, both 10/40), the command ip ospf network point-to-point, and that FULL/ - in show ip ospf neighbor means a link with no DR. Recognise 30/120 timers as the non-broadcast and point-to-multipoint types.

Network types at a glance

Test yourself on network types

Sources: Cisco OSPF FAQ: loopbacks and neighbour roles and Cisco show ip ospf interface reference. IOS-style outputs are illustrative; inspect actual media defaults and configured timers on your platform.