OSPF Fundamentals

OSPF is the interior routing protocol most enterprises actually run. It builds a shared map of the network and computes shortest paths from it, which makes it fast to converge and easy to reason about — once you understand areas and cost.

The idea

Every router floods information about its own links to every other router in the same area. Each router assembles an identical link-state database and runs Dijkstra's algorithm on it. Because everyone has the same map, everyone computes consistent paths.

Areas and the backbone

OSPF splits a large network into areas so that topology changes stay local.

• Area 0 is the backbone; every other area must connect to it.
• ABR (area border router) sits between area 0 and another area, and summarises between them.
• ASBR connects to a different routing domain and injects external routes.
• LSA types you will meet: 1 (router), 2 (network/DR), 3 (summary from ABR), 4 (ASBR summary), 5 (external), 7 (NSSA external).

graph TD
    A[Area 0 backbone] --- B[ABR]
    B --- C[Area 10]
    B --- D[Area 20]
    A --- E[ASBR]
    E --- F[External domain]

Single-area OSPF works for small networks and is a reasonable starting point. Multi-area exists to bound flooding and give you summarisation points.

Cost, and the trap in the default

Cost = reference bandwidth ÷ interface bandwidth. The reference defaults to 100 Mbps, so:

Every link at 100 Mbps or faster gets cost 1, so OSPF cannot tell a 1 Gbps link from a 40 Gbps one. The fix is to raise the reference bandwidth on every router, consistently:

# Cisco
router ospf 1
 auto-cost reference-bandwidth 100000     # 100 Gbps in Mbps

# FRR / Linux
ip ospf cost 10                            # or set cost per interface

Changing it on one router only is worse than not changing it: two routers now disagree about the map.

Neighbours, DR and BDR

On a broadcast segment OSPF elects a Designated Router and Backup DR to reduce adjacencies from n(n−1)/2 to 2n−1. Election is by interface priority (default 1), then router ID — highest wins. Priority 0 means "never".

Neighbours must agree on: area ID, hello/dead timers, subnet and mask, authentication, MTU (for DBD exchange), and stub flags. Any mismatch means the adjacency sticks in INIT, EXSTART or EXCHANGE, and routes never appear.

show ip ospf neighbor
show ip ospf interface brief
show ip route ospf
show ip ospf database

Where it goes wrong

• MTU mismatch. The adjacency forms and then breaks at database exchange, flapping every few minutes — a tunnel with a smaller MTU on one end is the classic cause.
• Reference bandwidth changed on one router. Asymmetric costs, traffic taking the slow path, and no error anywhere.
• Missing area 0. Two areas connected directly without the backbone create a virtual-link dependency that nobody documents.
• Summarisation blackholes. A summary that is wider than the addresses actually present silently discards traffic for the gap.
• Timers mismatched after a vendor migration. Hello 10/dead 40 on one side, 30/120 on the other: neighbours never come up, and the log says nothing useful.
• Redistribution loops. Mutual redistribution between OSPF and BGP (or static) without filters re-injects routes endlessly.
• Authentication off "temporarily". Anyone on the segment can inject LSAs; the temporary change outlives the reason.

Related Notes

• Routing Fundamentals — distance-vector vs link-state
• BGP And Internet Routing — the protocol OSPF hands off to at the edge
• Network Design And Topologies — where areas come from
• Route Selection And Metrics — what wins when protocols disagree