Basic Setup

Data Link Layer "Routing": Directly Connected

Only outgoing interface and L2 address (MAC) needed
L2 address obtained through ARP in IPv4 and NDP in IPv6
Frames are selectively flooded on L2
Switches use a lookup table to map L2 addresses to outgoing interfaces
If unknown, flood everywhere except source interface
STP takes care of loop, as to not create a broadcast storm with flooding

Global Routing: Not Directly Connected

When not directly connected, we need a gateway (next hop)
Use the system's default gateway: 0.0.0.0/0 -> eth0
Routing table:
Keep track of best paths
Forwarding: choose which path (there may be multiple) to use

Use recursive lookup
In practice only 1 level, so gateway should be directly connected
In theory, deeper recursion is possible

Classical (Outdated) Commands

Directly Connected IP Networks

Traditional config using ifconfig command
iwconfig for wireless interfaces

Routing Table

netstat -r: The traditional command to show the routing table
Contains:

• The destination
• The gateway (next hop)
• The subnet mask of the destination address
• The interface to forward it to

route is the same as netstat -r, but also allows modifying the route table
Add a static default route:
route add default <gateway>
Add a static host route:
route add -host <host> gw <gateway>
Add a static network route:
route add -net <network> netmask <netmask> gw <gateway>

You could consider adding static routes instead of letting the protocols work:

• Deterministic, always the same
• Backup options

ARP Table

Read the ARP table: arp
Delete an entry: arp -d <address>
Add a static entry: arp -s <address> <hardware_address>
This also prevents ARP spoofing
ARP data stays in the subnet, the router is the border

Modern Commands

The ip command

Introduced by linux iproute2 package
Meant to replace ifconfig, route and arp
Has support for:
IPv6
Multicast
Multiple routing tables (e.g. for VLANs)

Subcommands:
link: L2 settings (MAC)
address L2 settings (IP)
maddress: L2/3 multicast membership
neighbour ARP cache date
rule: routing table selection

ss, iw and ethtool commands

ss replaces netstat: shows socket status and statistics
iw replaces iwconfig: a utility for wireless devices
ethtool: shows and controls the network driver and hardware status

Route Selection

Longest prefix match: host routes are preferred over network routes
If you know the full address, might as well use it
Default route has shortest prefix
Last resort
Cannot be used in the internet core: might create loops, if present

Mathematical Representation

As in previous lectures: undirected graphs
Routes are nodes, connections are edges
Labels on edges represent cost
Broadcast networks are represented using "extra" nodes

Node Reduction

Lan with $N$ routers/hosts
Full mesh has $N(N-1)/2$ edges -> $O(N^2)$ complexity
With 1 extra node, it can be reduced to $O(N)$
Just an added fake node

Works for LANs, but also for Non-Broadcast Multiple Access (NBMA) networks
In many situations
But not always
Assumes all links are the same and working

Routing on the Internet

Based on top-level structure defined by Autonomous Systems (AS)
Each AS has control of a collection of:
- Routers and hosts
- Networks
Similar to the post office of a town
Can be operated by ISPs, companies, universities, ...
There are tiers to ASs
Tier 1: backbone to the internet, no upstream
Tier 2: can upstream to tier 1 AS
Tier 3: smaller ISPs

Edge routers in one AS can be directly connected to edge routers in another AS
Used for inter-AS routing
Implementing an Exterior Gateway Protocol (EGP)
Border Gateway Protocol 4 (BGP4) is the current choice
No other protocols in use, except variants of BGP

Routers within an AS communicate with one another to provide intra-AS routing
Implementing an Interior Gateway Protocol (IGP)
E.g. RIP, OSPF, IS-IS
Extended with an internal exterior protocol (iBGP): the internal part of an exterior protocol

Different Routing Mechanisms

Routing Protocol Classifications

Static:
A "human" protocol
Explicit commands like route add
Side effect of a script
Dynamic:
Automatically determined by a routing protocol
E.g.:
Distance vector (RIP)
Path vector (BGP)
Link state (OSPF, IS-IS)

Distance Vector Routing

Original ARPANET routing algorithm, based on Bellman-Ford
Decentralised
Asynchronous
No selection of "root"
Matrix with costs from node to node
In the end, every node will know the shortest destination from node to node

Path Vector Routing

Like distance vector routing, but
Instead of the distance to the destination, the complete path (on AS level) is specified
"It takes 3 hops to reach X" -> "To get to X you have to go through A, B and C"
Still decentralised and async
May depend on explicit or implicit policies
Has consequences for stability and convergence due to incompatible policy rules

Link State Routing

Link state algorithm
Dijkstra
Single source shortest path
Used by OSPF for intra-domain routing
Complete knowledge is distributed to all nodes
Every node executes the algorithm and hence draws consistent conclusions
Scales better than other options
Resolves loops

1. Distribute topology information
2. Run Dijkstra on the received topology