What

Exterior Gateway Protocol (EGP) for routing between ASes
Policy-based rather than metric based
Money makes the BGP go round

Why

Allows the use of policies, enabling ASes to make deals with one another on routing
Path vector algorithm: prevents loops
Flexible:
Complex policies
Traffic engineering
Multi-homing
Load balancing

Concepts

Autonomous System (AS)
An administrative domain with a unified routing policy, e.g. an ISP or large company
Each AS has a unique AS Number
BGP sessions are on port 179 using TCP

BGP types
External BGP (eBGP): used to route between ASes
Internal BGP (iBGP): within one AS

Communities
A way to communicate preferred treatment
The meaning of certain numbers has to be agreed upon beforehand between 2 or more parties
When a route is forwarded to a party that does not have an agreement on community meanings, the community field is emptied

Multi-Exit-Discriminator (MED)
- Used to tell peer which entry point to prefer
- Lower MED = better
- Influences incoming traffic
- Only a suggestion

Tables

• Neighbour Table: BGP peers and their state
• BGP Table (Adj-RIB-In): A table per peer containing routes received before filtering
• BGP Table (Loc-RIB): Best routes selected by BGP decision process
• BGP Table (Adj-RIB-Out): A table per peer containing routes advertised to said neighbour after filtering
• Routing Table (RIB): Best BGP routes installed for forwarding (if better than IGP routes)

Attributes

• ORIGIN: How route entered BGP (IGP, EGP, Incomplete)
• AS_PATH: List of ASes route has traversed (loop prevention, path selection)
• NEXT_HOP: IP address of next router to reach destination
• LOCAL_PREF: Preference within AS (higher = better), not sent to eBGP peers
• MED (Multi-Exit Discriminator): Suggestion to external AS about preferred entry point (lower = better)
• COMMUNITY: Tags for grouping routes and applying policies
• WEIGHT: Cisco-proprietary, local to router only (higher = better)

Neighbour States

• Idle: Initial state, waiting to start
• Connect: TCP connection initiated
• Active: Trying to establish TCP connection (despite name, this indicates problems)
• OpenSent: TCP established, Open message sent
• OpenConfirm: Open message received, waiting for Keepalive
• Established: Session up, exchanging routes

Messages

• Open: Establishes session, negotiates parameters (AS number, Router ID, hold time)
• Update: Advertises new routes or withdraws unreachable routes
• Keepalive: Maintains session (sent every 60s by default)
• Notification: Reports errors, closes session

(S)Election

All BGP speakers are equal, but there is path selection
The following attributes are used:

• Highest WEIGHT (Cisco-proprietary, local to router)
• Highest LOCAL_PREF (prefer certain exit paths within AS)
• (isn't this done through LOCAL_PREF?) Locally originated (prefer own routes)
• Shortest AS_PATH (fewer ASes to traverse)
• Lowest ORIGIN (IGP < EGP < Incomplete, mostly 0 nowadays)
• Lowest MED (if from same AS, prefer suggested entry)
• eBGP over iBGP (prefer external routes)
• Lowest IGP metric to NEXT_HOP (closest exit point)
• Oldest route (stability preference)
• Lowest Router ID (tiebreaker)

How

1. Configure BGP on your router
   • Set AS number
   • Set router ID
   • Define peers
   • Advertise certain network(s)
2. TCP connection is initiated to peers (3-way handshake)
3. R1 send BGP OPEN message
4. R2 receives, validates and if the params are acceptable sends an OPEN back to R1
5. R1 receives and if the params are acceptable sends a KEEPALIVE (goes to OpenConfirm state)
6. Once KEEPALIVEs have been exchanged, both routers go to Established state
7. Send BGP UPDATE messages to exchange routes
8. Receive updates and process them, if acceptable store them and populate the routing table