Interior Routing — OSPF

OSPF (Open Shortest Path First) is the interior gateway protocol most organizations actually run. It is a link-state protocol: every router describes its own directly-attached links, floods that description to every other router in the area, and from the flood each router assembles an identical map of the whole topology. Then each router independently runs Dijkstra's shortest-path-first algorithm over that map to compute its own best path to every destination.

Because every router shares one consistent map, OSPF converges fast — sub-second on a well-built network — and is structurally immune to the counting-to-infinity loops that plague rumor-based protocols. It scales comfortably across a single organization's network. What it does not do is scale to the internet: flooding every router's link state to everyone is fine for thousands of links and impossible for the millions of prefixes BGP carries.

describe
own links

→

flood
LSAs

→

identical
LSDB

→

run
Dijkstra

Link-State versus Distance-Vector

The dividing line in interior routing is what each router knows. A distance-vector protocol (RIP, EIGRP) knows only what its neighbors tell it — "I can reach network X at cost N" — and never sees the topology itself. It routes on hearsay, which is simple but slow to converge and historically prone to loops, because a router can believe a stale rumor long after the underlying link failed.

A link-state protocol gives every router the full map. Instead of trusting a neighbor's summary, each router learns the raw link facts from everyone and computes paths itself. The result converges faster and cannot loop the way distance-vector can, because no router is acting on a second-hand summary it cannot verify. The price is memory and CPU: holding the whole map and recomputing Dijkstra on every change costs more than tallying a few neighbor reports.

LSAs and the Link-State Database

The unit OSPF floods is the LSA (Link-State Advertisement): a small record describing one router's links, their neighbors, and their costs. When anything changes — a link goes up or down — the affected router originates a fresh LSA and floods it reliably to every router in the area. Each router stores every LSA it receives in its link-state database (LSDB), and OSPF's core guarantee is that every router in an area holds an identical LSDB.

That identical database is the whole trick. Routers first form adjacencies with neighbors, synchronize their databases, and only then compute routes — so two routers that fail to fully synchronize never become adjacent and never exchange routes. A subtle but common cause of a stuck adjacency is an MTU mismatch: the database-description packets are sized to the interface MTU, and if the two ends disagree, the larger packet is dropped and the adjacency hangs in EXSTART forever.

# inspect OSPF neighbors and the synchronized database (FRR / Quagga)
show ip ospf neighbor
# Neighbor ID   State    Interface     <- FULL = synced; EXSTART = MTU mismatch
# 10.0.0.2      Full     eth0
# 10.0.0.3      ExStart  eth1          <- stuck: check MTU on both ends

show ip ospf database
# the LSDB — identical on every router in the area, by design

Dijkstra and Cost

With an identical map in hand, each router runs Dijkstra's shortest-path-first algorithm to find the lowest-total-cost path to every destination. The metric OSPF minimizes is cost, an integer assigned per link and typically derived from bandwidth — a faster link gets a lower cost, so Dijkstra naturally prefers it. The path with the smallest summed cost wins; hop count never enters the calculation.

Because cost drives everything, a single mis-assigned cost quietly bends traffic the wrong way. Default cost is computed from a reference bandwidth, and on modern networks every link faster than that reference collapses to the same minimum cost unless you raise the reference — so a 10 Gbps and a 1 Gbps link can look equal to Dijkstra, and traffic takes the slower one half the time. Setting cost deliberately, rather than trusting the defaults, is how you control OSPF path selection.

Areas

Flooding has a natural ceiling: every change ripples to every router, so a large flat OSPF domain spends more and more on flooding and recomputation as it grows. OSPF's answer is areas — partitions that bound the flood. LSAs flood only within their area; the backbone area 0 ties the areas together, and routes between areas are summarized rather than flooded in full detail.

Areas turn one giant link-state database into several smaller ones, so a flapping link in one area never forces a full SPF recomputation across the entire network. This is what lets OSPF scale within an organization — but it is also why one undisciplined area-0-only design eventually buckles: with no segmentation, every router holds every LSA and recomputes on every change anywhere.