OSPF network types: broadcast, point-to-point, NBMA, point-to-multipoint and loopback

Why network types exist

OSPF was designed when networks used many kinds of links: Ethernet, serial lines, and multi-access WAN clouds such as Frame Relay that could reach many sites but could not send broadcasts. Each kind needs different handling. The network type setting tells OSPF which handling to use on an interface. A router picks a default from the interface type, and you can change it with ip ospf network.

The types compared

Because hello and dead timers must match, two routers using types with different timers (for example broadcast and point-to-multipoint) will not become neighbours. Two types with the same timers but different DR behaviour may form an adjacency yet build an inconsistent database, so routes are missing. The safe rule is: set the same network type at both ends.

Choosing the right type

• Ethernet link between exactly two routers: change it to point-to-point. There is no DR election or wait timer, adjacencies form faster and the database is smaller because no type 2 LSA is needed.
• Ethernet segment with several routers: leave it as broadcast and control the DR with priorities.
• Hub-and-spoke overlays such as DMVPN: tunnel interfaces default to point-to-point, which allows only one neighbour, so the hub must change it. point-to-multipoint is the simplest choice and needs no DR; broadcast keeps spoke-to-spoke next hops but needs the hub to be DR (spokes priority 0). Spokes must match the hub's choice.
• Non-broadcast and point-to-multipoint non-broadcast were built for Frame Relay and ATM, legacy WAN services now replaced by MPLS, Ethernet services and internet VPNs. You mainly meet them in exams and old networks.

Next hops and loopbacks

Network types also change the next hop other routers install. On broadcast and non-broadcast types, OSPF keeps the real next hop of the router that owns the destination, even when traffic could pass through the hub. That works only if every router can reach every other directly at Layer 2. Point-to-multipoint treats the cloud as a collection of separate links and advertises a /32 host route for each router, so traffic goes through the hub. That is slower but always reachable.

A loopback interface is always advertised as a /32 host route, whatever mask you configure, because OSPF knows only one address can live there. If a lab or a service needs the real subnet advertised, set ip ospf network point-to-point on the loopback.

interface GigabitEthernet0/2
 description Link to R2 only
 ip ospf network point-to-point
!
interface Tunnel0
 description DMVPN hub
 ip ospf network point-to-multipoint
!
interface Loopback10
 ip address 172.16.10.1 255.255.255.0
 ip ospf network point-to-point

Real-world example: slow failover on a campus core

A college ran OSPF between two core switches and four distribution switches, with each link a separate routed Ethernet port. After a distribution switch reboot, users waited almost a minute for routing to return. Every link was still the default broadcast type, so after each reboot the routers waited out the 40-second DR wait timer before forming adjacencies, and the database held a network LSA for every two-router link. Setting ip ospf network point-to-point on both ends of every core and distribution link removed the elections. Adjacencies then came up within a couple of seconds of the link, and the LSDB became smaller and easier to read.