71. OSPF in depth

71.1 The model

Open Shortest Path First (RFC 2328 for OSPFv2/IPv4; RFC 5340 for OSPFv3/IPv6) is the dominant enterprise link-state IGP.

1. Discover neighbours    (Hello packets, multicast 224.0.0.5)
2. Form ADJACENCIES       (not with everyone -- see DR/BDR)
3. Exchange the LINK-STATE DATABASE (LSDB) -- every router ends up with an
   IDENTICAL copy for its area
4. Run DIJKSTRA (SPF) on that database, from itself, to get shortest paths
5. Install the results in the routing table
6. Re-flood on any change; re-run SPF

OSPF runs directly on IP, protocol number 89 — not TCP or UDP. It provides its own reliability (acknowledged LSAs).

71.2 Neighbour states and adjacency

Down -> Init -> 2-Way -> ExStart -> Exchange -> Loading -> FULL
                  ^                                          ^
          "I see you seeing me"              databases synchronised

Stuck states are diagnostic:

Stuck at	Almost always means
`Init`	Hellos are one-way — the other side isn't hearing you (ACL, unidirectional link)
`2-Way` (on a broadcast segment)	Normal — non-DR routers stay 2-Way with each other by design
`ExStart`	MTU mismatch between neighbours (Ch. 15 again)
`Loading`	Packet loss or LSA problems
Flapping	Hello/dead timer mismatch, or a genuinely unstable link

Hello/dead intervals must match (default 10 s / 40 s on broadcast links), as must the area ID, authentication, and the stub-area flag, or the adjacency never forms.

71.3 DR and BDR

On a broadcast segment with n routers, full mesh adjacency would need n(n-1)/2 relationships. OSPF elects a Designated Router and Backup DR; everyone else forms full adjacency only with the DR/BDR and stays 2-Way with each other.

Election: highest OSPF priority (default 1; 0 = never DR), tie-break on
          highest Router ID.
          NOT pre-emptive -- a better router arriving later does NOT take over.

DR/BDR does not exist on point-to-point links. On point-to-multipoint or NBMA topologies, misconfigured network types are a classic source of "adjacency won't form".

71.4 Areas and LSA types

OSPF scales by hierarchy. Every area must connect to area 0 (the backbone); inter-area traffic transits area 0.

        +------------------ Area 0 (backbone) ------------------+
        |                                                       |
   [ABR]|                                                       |[ABR]
        |                                                       |
   Area 1                                                    Area 2
   (SPF runs only within an area; ABRs summarise between them)

LSA type	Name	Carries	Scope
1	Router LSA	This router's own links and costs	within the area
2	Network LSA	The segment, generated by the DR	within the area
3	Summary LSA	Inter-area prefixes, generated by an ABR	into another area
4	ASBR Summary	How to reach an ASBR	between areas
5	External LSA	Routes redistributed from outside OSPF, by an ASBR	flooded domain-wide
7	NSSA External	Externals inside a not-so-stubby area (converted to type 5 at the ABR)	within an NSSA

Area types control which LSAs are allowed in, trading visibility for size:

Area type	Blocks	Result
Standard	nothing	Full LSDB
Stub	type 5 (externals)	ABR injects a default route instead
Totally stubby	types 3, 4, 5	Only a default route — smallest tables
NSSA	type 5, but allows type 7	A stub area that still has its own external connection

Roles: ABR = Area Border Router (touches area 0 and another area); ASBR = Autonomous System Boundary Router (redistributes external routes in).

71.5 Cost, and the reference-bandwidth trap

cost = reference_bandwidth / interface_bandwidth

Default reference bandwidth = 100 Mbit/s. Therefore:
   100 Mbit link  -> cost 1
   1 Gbit link    -> cost 1     <-- same!
   10 Gbit link   -> cost 1     <-- same!
   100 Gbit link  -> cost 1     <-- same!

Every link at or above 100 Mbit/s has cost 1 by default, so OSPF cannot tell a gigabit link from a 100-gigabit one, and picks by hop count instead. Fix: auto-cost reference-bandwidth 100000 (100 Gbit/s) — and set it identically on every router, or you get asymmetric routing.

Equal-cost paths are load-balanced (ECMP, Ch. 50.3).

71.6 OSPFv3 differences

Runs over IPv6 (protocol 89 still), addresses in LSAs are separated from topology, so it can carry multiple address families.
Adjacencies form over link-local addresses (fe80::).
Authentication is delegated to IPsec rather than being built in.
Router IDs are still 32-bit dotted-quad values, even in IPv6.

71.7 Real-world example

A campus network had two data centres connected by both a 10 Gbit/s dark-fibre link (2 hops) and a 1 Gbit/s backup (1 hop). Traffic consistently chose the 1 Gbit/s path. Cause: default reference bandwidth, so both links cost 1, and the 1-hop path had a lower total cost. Raising the reference bandwidth to 100 Gbit/s on every router made the fibre cost 10 and the backup cost 100 — traffic moved immediately. Check reference bandwidth on every OSPF network you inherit.

Try it yourself

# FRRouting on two Linux boxes/containers
sudo vtysh
  configure terminal
   router ospf
    ospf router-id 1.1.1.1
    auto-cost reference-bandwidth 100000
    network 10.0.0.0/24 area 0
   exit
  end
  show ip ospf neighbor          # look for state FULL
  show ip ospf database          # the LSDB -- LSA types visible
  show ip ospf interface eth0    # cost, hello/dead, DR/BDR, network type
  show ip route ospf