OSPF Inter-Area Routing

Inside an area every router holds the whole map, but no router LSA ever leaves its area. Routes between areas are carried by the ABRs as summary LSAs: a prefix and a cost, with nothing of the links behind them. This demo follows one such route from area 2 to area 1 across the network from the Hello , Database Exchange and Shortest Path First demos, then configures an area range to show what summarising addresses changes.

Areas and ABRs

• Each area has its own database. R1A holds area 1's router LSAs and nothing else, so it knows every link in area 1 and none beyond.
• An ABR sits in two areas. ABR-2 holds area 0's database and area 2's, and runs SPF on each separately.
• The ABR carries the routes across. From its SPF in area 2, ABR-2 knows R2B's LAN, 10.2.100.0/24, costs 10 + 10 + 10 = 30 through R2A, and says so to area 0.

Summary LSAs: prefixes without the links

An ABR tells the next area every prefix it can reach, one summary LSA (type 3) each, carrying the prefix and the ABR's cost to it:

• The links stay behind. Area 2's database says which router connects to which, over which link, at what cost. Area 0 gets only the four prefixes and ABR-2's cost to each.
• No address is left out or merged. Area 0 receives less data, but it still sees every one of area 2's networks. That is not summarisation in the configuration sense: the name "summary LSA" means the topology is summed up as a cost, not that the addresses are.
• The prefixes are summarised only when an area range is configured, the subject of the last section.

Distance vector between areas

Across an ABR, OSPF works the way a distance-vector protocol such as RIP does: a router trusts the cost it is given and adds its own.

• ABR-1A and ABR-1B each reach ABR-2 at 10 across the area-0 segment, so each reaches 10.2.100.0/24 at 10 + 30 = 40.
• Each sends its own summary LSA into area 1, at 40. Every area-1 router now holds two for the prefix, one per ABR.
• An area-1 router adds its own cost to each ABR, from its SPF in area 1, and takes the lower sum.

Area 0 as the hub

A distance vector can loop: a router taking a neighbor's cost cannot tell whether that cost was worked out through itself. OSPF rules this out by restricting where summaries may go.

• Summaries go from an area into area 0, and from area 0 out to the other areas. An ABR sends area 0 only the routes of its own areas, never the ones it learned from area 0.
• An ABR uses only the summaries it hears in area 0. ABR-1A also hears ABR-1B's summary of 10.2.100.0/24 inside area 1, and ignores it.
• The areas therefore form a star around area 0. A route passes through area 0 once and cannot come back to where it started. This is also why every area must touch area 0, directly or through a virtual link.

Choosing an exit from area 1

Area 1 has two ABRs, and the same two summaries give different routers different answers:

• Each router leaves by the exit nearest itself. Beyond the ABR, the packet follows that ABR's table.
• A tie keeps both exits. R1C is 25 from each ABR, so it shares the traffic between them.
• The same happens in the other direction. ABR-1A summarises R1A's LAN, 10.1.100.0/24, into area 0 at 25, and ABR-1B at 30; ABR-2 adds 10 to each and takes ABR-1A at 35.
• An intra-area route always wins. A router prefers a path inside its own area to one through another, whatever their costs; routes from summaries show as O IA in its table.

Area ranges: summarising addresses

An area range makes the ABR send one summary LSA for a whole block of prefixes instead of one per prefix. With area 1 range 10.1.0.0/16 configured on both area-1 ABRs:

• Area 0 holds 2 summaries for area 1 instead of 18. All nine of area 1's prefixes fall inside 10.1.0.0/16, and each ABR now advertises the range alone.
• The range costs as much as the most distant network inside it. RFC 2328 sets its cost to the largest of the costs it covers: 45 from ABR-1A (the direct R1A–R1D link, 10.1.7.0/30) and 40 from ABR-1B (the same link).
• Routing outside area 1 gets coarser. ABR-2 compares 10 + 45 with 10 + 40 and sends everything for 10.1.0.0/16 through ABR-1B. Traffic to R1A's LAN now costs 10 + 30 = 40, five more than through ABR-1A: the range hides which ABR is nearer to each network.
• A change inside area 1 stays inside. When a link in area 1 fails, the range is advertised again only if its largest cost changes; without the range, every prefix whose cost moved would be.

Simplifications in the demo

• Only type-3 summaries appear: there are no external routes, so no ASBR and no type-4 or type-5 LSAs.
• The area range is configured on both ABRs at once. Most implementations also install a discard route for the range on the ABR, so traffic to an unused address inside it is dropped there; the demo leaves that out.
• Ties go to both next hops, as OSPF does; the drawing shows one of them.