• Routers communicate with neighbour routers advertising networks as measures of distance and vector 
• Mesures:
• Distance = Metric (Hop count - how many routers to arrive)
• Vector = Direction (Which Interface you're going to go out of)
• Knowledge of the network is passed from directly connected neighbours
• Send periodic updates of the entire routing table

• Routers communicate with all other routers exchanging link state information to build a topology of the entire network
• Measures:
• Link-state = interface connections or "links" to other routers and networks 
• Routers have a complete view of the network, knowledge of the entire topology
• Send triggered partial updates

• Distance = Metric (Hop count - how many routers to arrive)
• Vector = Direction (Which Interface you're going to go out of)

Link-state = interface connections or "links" to other routers and networks

• Simple, flat design, non-hierarchical networks
• minimum administrator knowledge
• Convergance time is not an issue

• Large, hierarchical networks
• Advanced administrator knowledge
• Convergance time is crucial

The level of 'awareness' of routers of the events happening in the topology they belong or are aware of.

This is achieved by having the networks communicating with each other, notifying each other about a certain 'happening' in the network

Eg. if a network goes down, once all the nodes have known that the network has gone down, the network has converged.

The former does not send subnet mask information with their routing updates reacting in the following two ways:

• If the router has a directly connected interface belonging to the same major network it will apply the same subnet mask as the interface
• If the router does not have any interfaces belonging to the same major network, it will apply the classful subnet mask to the route

Belonging to the same major network just means they belong to the same classful network:

• 10.3.1.0 and 10.5.5.0 would belong to 10.0.0.0

The latter does send the subnet mask with their updates, thus , variable length subnet masks (VLSMs) are allowed when using classless routing protocols

• Knowledge of next hop and inteface to use
• Exchange best route (shortest distance) information for any known prefixes only with direct neighbours

• Use full knowledge of routers/topology in site
• Routers flood information messages describing their connected neighbours around the entire site network

• It is based on Bellman-Ford algorithm
• Later Updated to RIPv2
• Added support for authentication for CIDR prefix
• RIPng defined to add IPv6 support
• Uses distance vector algorithm
• Router sends whole routing table periodically (Every 30 secs) in messages to directly connected routers
• Destination network (prefix) and distance (cost) in hops
• Receiving routers use messages to update their view of the best route (lowest distance) to a given network (prefix)

• Each router maintains a routing table
• Destination network 
• Cost (hop count); 0 if network directly connected
• Next hop to use for best path
• If a router receives an update with a lower cost path to a destination network it updates its entry for that destination
• Message propagation takes time (30 secs per update)
• If it receives higher cost path later form the same neighbour, the higher value is used
• Because topologies can change

• Destination network
• Cost (Hop count); 0 if network is directly connected)
• Next hop to use for best path

Updates its entry for that destination

(Message propagation takes time - 30 secs per update)

• Updates only sent every 30 seconds
• Updates are not acknowledged
• Metrics are simple hop count values
• Limited to max value of 15; 16 = unreachable
• Suitable for smaller homogeneus networks
• No detail weighting - just hops
• Routers don't have knowledge of network topology
• Can be slow to converge when problems occur

If we have a subnet connected like A-B-C-D-E-F, and the node A goes down, B does not receive the vector update from A. THe problem is that B gets an update from C saying that A is 2 hops from itself, as C did not receive any update about A being down. This slowly propagates to the network until it reaches infinity.

It can be solved with Split Horizon Heuristics - don't advertise a route on the interface it was learnt from.

• It says don't advertise a route on the interface it was learnt from
• Requires extra state to be held for the route
• Solves simpler "Count to infinity" problems
• Non-existing routes no longer advertised
• Dead links will time out and be removed

• Discover neighbours
• Determine cost metric to each neighbour
• Construct link state information packet
• Flood this message to all site routers in same area
• Use messages to build topology and then compute shortest paths for prefixes served by any given router
• All routers learn the full topology

• Send a broadcast message on all interfaces
• Adjancencies built by received "Hello" messages
• Then measure or manually configure line cost
• Typically bsed on bandwith/delay

• Source ID to uniquely identify the node
• Sequence to allow receiver determine if message is new one to process and flood
• All message are ACK to senders
• Age
• Decrement one per second
• Prevents old messages persisting in the network

• Sender increments Sequence number with each message
• Floods on all links except one received on
• Messages buffered for a short time for robustness

• Once all routers have the topology we need to find shortest path
• Uses Dijkstra's algorithm
• Determines shortest path through a graph from an initial node to any given destination
• Algorithm works by exapanding from the starting node, considering cheapest neighbour with each iteration
• Thus the algorithm finds the next "closest" node with each iteration, until it reachest the destination

1. Set tentative cost to reach all nodes from initial node to infinity, and cost of reaching initial node to 0
2. Mark all nodes unvisited
3. For the current node
1. For all unvisited neighbours of current node
1. If cost via link from current node to neighbour is less than existing tentative cost, update cost
2. Mark the current node as visited
3. If destination node is visited, stop
4. Else make the node with the lowest cost the current node and repeat the iteration
4. Shortest path can be found by following the reverse path

• Link state oriented
• v2 for IPv4 and v3 for IPv6
• Supports concept of "areas"
• Allows a site to be split for purposes of routing
• Each router interface can be in only one area
• Sends updates via multicast
• Set of link state relationships referred to as an adjacency database

• Inter-site routing uses exterior routing protocols
• You may advertise your network prefixes to neighbouring networks
• You may or may not offer transit to other networks
• Policy is often more important than path costs
• Border Gateway Protocol (BGP)

• A protocol for exchanging routing information between gateway hosts in a network of Autonomous Systems. 
• Works between Autonomous Systems
• Each ISP has a unique ASN assigned by RIRs
• Routing in BGP includes information about the AS path a given route
• Routing table contains a list of known routers, the addresses they can reach, and a cost metric associated with the path to each router so that the best available route is chosen

• In configuration, specify IP of neighbour and AS
• e.g. neighbour 152.40.30.20 remote-as 3456
• Creates BGP peering session (TCP, port 179)
• Initially sends whole routing table
• Then incremental UPDATEs
• Advertise routes-to-other-networks to neighbours
• Network layer reachability Information messages (NLRI)
• Contains prefix, prefix length, AS path, hop
• Neighbour then chooses whether to use route
• BGP is loosely distance vector but includes many improvements
• e.g. know full path, so can detect loops if your own AS on path

• Border gateway protocol is "routing with politics"
• Wide range of attributes to base decisions on:
• Weight (path cost)
• Origin of route, how route was learnt
• AS path (list of intermediate AS to destination)
• Community tag
• ...
• BGP communities
• Allows policy to be described for "communities" of networks
• e.g. ask neighbour to not advertise certain routes further
• Helps control who provides transit to 
• Which ISPs carry traffic from a given AS