Dynamic Routing

1Recap4 min

Today we fill the table differently. The new sources of routes - RIP, OSPF, EIGRP - each have their own administrative distance and their own metric, and the router must compare them all. The order of comparison is exactly the one from lecture 6, and it is worth having fresh in mind.

Learning outcomes

• Explain why static routes do not scale and what exactly a dynamic protocol solves
• Distinguish between distance-vector, link-state and path-vector
• Describe the loop problem in distance-vector and the mechanisms that mitigate it
• List the four steps of a link-state protocol
• Apply Dijkstra to a small topology and obtain the shortest-path tree
• Calculate the OSPF cost of a link and know when the formula gives wrong results
• Explain why a DR and a BDR are elected and how the election proceeds
• Configure single-area OSPF and troubleshoot an adjacency that will not form

2Why static routes are not enough7 min

The two do not exclude each other. Even in an entirely OSPF network, the default route towards the provider is almost always static - because there is nobody to negotiate it with.

3The families of protocol8 min

The first division concerns where they are used: inside an organisation, or between organisations.

The second division, the one that concerns us today, is about what each router knows:

4Distance-vector and the problem of loops10 min

A distance-vector protocol such as RIP works on a simple principle: each router periodically sends its neighbours its entire routing table, and they add the cost of the link and decide. Nobody sees the topology; each believes what it is told. Hence the nickname that stuck to it for good: routing by rumour.

Counting to infinity

5How a link-state protocol works10 min

The central idea is that each router learns from its neighbours not the routes but the state of the links throughout the network, builds its own complete map, and computes the paths for itself. Four elements make this possible.

Adjacencies and the Hello protocol

Two directly connected neighbours form an adjacency. The relationship is established and maintained through periodic Hello messages, which have a dual role: keep-alive - if the messages stop, the neighbour is considered down - and parameter synchronisation. Known neighbours are kept in a neighbour table.

ID_R2 - Fa0/1 - 00:13 - FULL
ID_R3 - Fa0/2 - 00:00 - LOADING
!       interface  time since last hello  adjacency state

The LSP: the unit of information

An LSP (Link State Packet) describes a router's links: for each interface, the attached network, the kind of link, the cost, its state and the neighbour at the far end. LSPs are sent:

• when the routing process is initialised;
• as soon as a change appears in the topology - the triggered update mechanism, the source of fast convergence;
• periodically, at long intervals (in OSPF, once every 30 minutes), as a refresh.

As soon as a router receives a new LSP, it retransmits it to all its other neighbours. The process is called flooding and ensures that, within a few tenths of a second, the whole network has the same information.

The four steps

1. Adjacencies and connected networks. The router discovers its neighbours and its directly connected networks.
2. Flooding of LSPs. Each router announces what it knows and retransmits what it receives.
3. Populating the topology table. All the networks received are kept, with their costs - not only the best ones. This is the map.
4. Dijkstra. The shortest-path algorithm is run, with the current router at the top. Only the results reach the routing table.

6Dijkstra's algorithm11 min

Choose which router does the computing and notice that the tree obtained is different for each one: every router places itself at the top. The map is the same; the tree is not.

Then change the cost of a link in the tree and watch the traffic move to another path - exactly what happens in a real network when an administrator adjusts the declared bandwidth of an interface. It is the principal instrument of traffic engineering in OSPF: you do not change the cables, you change the costs.

7OSPF: the basic characteristics9 min

OSPF (Open Shortest Path First) was developed by the IETF from 1988 onwards, as an open alternative to the proprietary protocols of the day. OSPFv2 (RFC 2328) is the IPv4 version, OSPFv3 (RFC 5340) the IPv6 one.

8The messages and the adjacency states9 min

9DR and BDR on multi-access networks10 min

On a point-to-point link there is only one possible adjacency and no problem at all. On an Ethernet segment with n routers, if each formed an adjacency with each, there would be n(n−1)/2 relationships - 45 for ten routers, each with its own Hello messages, its own synchronisations and its own retransmissions.

The solution is centralisation: a DR (Designated Router) is elected, which receives and redistributes the updates, and a BDR (Backup DR), which takes over if the DR fails. All the other routers (DROTHER) form an adjacency only with the DR and the BDR. The number of adjacencies falls to 2(n−1).

The election criteria

1. The highest priority on the interface (1 by default). The value 0 excludes the router from the election.
2. On a tie, the highest Router ID.

The Router ID is itself determined in three steps, in strict order:

1. the value configured by hand with router-id;
2. the highest IP address on a loopback interface;
3. the highest IP address on an active physical interface.

! the router that must be the DR
R1(config)# interface gigabitEthernet 0/0
R1(config-if)# ip ospf priority 100

! the router that must never be DR or BDR
R3(config)# interface gigabitEthernet 0/0
R3(config-if)# ip ospf priority 0

! a stable Router ID, through an interface that does not go down
R1(config)# interface loopback 0
R1(config-if)# ip address 1.1.1.1 255.255.255.255
R1(config)# router ospf 1
R1(config-router)# router-id 1.1.1.1

! the election is not pre-emptive - it must be forced
R1# clear ip ospf process

10OSPF across several areas10 min

Dijkstra runs the more slowly the larger the graph, and any change in one corner of the network forces every router to recompute everything. The scaling solution is to divide the OSPF domain into areas: each area runs Dijkstra on its own graph, and for destinations outside it merely adds the distance to the border router.

The types of LSA

The codes in the routing table

When both an E1 and an E2 route exist to the same destination, E1 is preferred: being cumulative, it reflects the real cost of the path more faithfully. E2 is the default precisely because it is simpler - but it is also less accurate.

11Configuration and verification9 min

R1(config)# router ospf 1
R1(config-router)# router-id 1.1.1.1

! which interfaces take part - the mask is a WILDCARD, as with ACLs
R1(config-router)# network 192.168.10.0 0.0.0.255 area 0
R1(config-router)# network 10.0.0.0 0.0.0.3 area 0

! the interface facing hosts: announce the network but send no Hellos
R1(config-router)# passive-interface gigabitEthernet 0/0

! adjusting the reference - THE SAME value on every router
R1(config-router)# auto-cost reference-bandwidth 10000

! the default route to the provider, announced across the OSPF domain
R1(config)# ip route 0.0.0.0 0.0.0.0 203.0.113.1
R1(config-router)# default-information originate

Look at the first command: the third line shows 2WAY/DROTHER with priority 0. It is not an error - it is exactly what should happen between two ordinary routers on a multi-access segment. Only the adjacencies with the DR and BDR reach FULL.

And in the routing table the three origins can be read at a glance: O from the same area with cost 2, O IA from another area with cost 12, and O E2 external, with the fixed cost of 20 that does not change however long the internal path may be.

12Common mistakes4 min

• "The adjacency is stuck in Exchange" The two ends have different MTUs and refuse to synchronise their databases. show ip ospf interface on both, compare the MTU. Or ip ospf mtu-ignore, as an emergency measure.
• "The neighbours are in Two-Way and go no further" If they are two DROTHERs on a multi-access segment, this is the correct behaviour. Check who the DR is. Full adjacencies exist only with the DR and BDR.
• "I set reference-bandwidth on one router only" The costs become incoherent: the same path has different values seen from different ends. The same value on every router in the domain. No exceptions.
• "I used the network mask in the network command" OSPF requires a wildcard, exactly as ACLs do. network 10.0.0.0 0.0.0.3 area 0, not 255.255.255.252.
• "I forgot passive-interface on the segments with hosts" The router sends Hellos towards users: wasted bandwidth and an open avenue of attack - anybody can start a bogus OSPF process. passive-interface default, then no passive-interface only where neighbours genuinely exist.
• "The new router does not become DR, although it has priority 200" The election is not pre-emptive. It is not a fault. clear ip ospf process, if you really do need a re-election.
• "I configured OSPF, but the default route does not reach the others" A static route does not enter OSPF by itself. default-information originate on the edge router.

13Summary and glossary4 min

14Self-check questions6 min

15Further reading2 min

The next lecture leaves the wire and takes to the air: wireless networks, with their particular medium-access problems - you cannot listen and transmit at the same time, so CSMA/CD does not work - and with a security problem the wired network does not have: the medium is public by definition.

Laboratory 5 configures single-area OSPF on exactly the same topology on which you earlier wrote static routes. The comparison between the two configurations, in number of lines and in reaction to an unplugged cable, is the point of the exercise.

• RFC 2328 - OSPF version 2
• RFC 5340 - OSPF for IPv6
• RFC 3630 - traffic engineering extensions for OSPF
• RFC 2453 - RIP version 2, for comparison
• John Moy, OSPF: Anatomy of an Internet Routing Protocol