BGP: Border Gateway Protocol

Learning objective

Explain how BGP establishes peering sessions, advertises path attributes, and applies the RFC 4271 decision process to choose a best path.

What BGP is

BGP (Border Gateway Protocol) is the routing protocol that holds the public Internet together, standardized in RFC 4271 (BGP-4). Unlike the three interior gateway protocols in this course, BGP is an Exterior Gateway Protocol: it runs between independently administered networks called autonomous systems (AS), each identified by an AS number. BGP is a path-vector protocol — related to distance-vector, but instead of a simple distance it exchanges the entire sequence of autonomous systems a route has crossed, letting a router detect and reject a route that loops back through its own AS. Two BGP routers that exchange routes are peers; a session between routers in different autonomous systems is eBGP, and a session between routers in the same autonomous system is iBGP. BGP runs over TCP port 179, inheriting reliable, ordered delivery instead of implementing its own.

BGP message types

RFC 4271 defines four BGP message types. OPEN is sent once, right after the TCP connection forms, to negotiate session parameters (AS number, hold time, BGP Identifier). UPDATE advertises new reachable prefixes or withdraws ones that are no longer reachable, each carrying its path attributes. NOTIFICATION reports an error and closes the session. KEEPALIVE is sent periodically (and has no body beyond its header) to confirm the session is still alive within the negotiated hold time.

The BGP finite state machine

A BGP session moves through a defined sequence of states: Idle (no connection attempted) → Connect (waiting for the TCP connection) → Active (retrying the TCP connection) → OpenSent (TCP is up, local OPEN sent, waiting for the peer's OPEN) → OpenConfirm (OPEN messages exchanged, waiting for a KEEPALIVE) → Established (the session is fully up). Routes are only exchanged once a session reaches Established — anything short of that is still session setup, not routing.

Path attributes

Every BGP UPDATE carries path attributes describing the route, which the decision process later compares. AS_PATH lists every autonomous system the route has crossed; NEXT_HOP is the IP address to forward toward; ORIGIN records how the route entered BGP (IGP, EGP, or Incomplete); LOCAL_PREF is an iBGP-only attribute an administrator sets to prefer one exit point over another; MED (Multi-Exit Discriminator) is a hint one AS gives another about which of several entry points it prefers, deliberately weak in the decision process so it rarely overrides more important attributes.

Interactive BGP session establishment

Step through two eBGP peers moving through the finite state machine from Idle to Established, then advertising a prefix with its path attributes.

The BGP decision process

When more than one route to the same prefix survives, RFC 4271's decision process compares them in order, stopping as soon as one criterion breaks the tie: highest LOCAL_PREF, shortest AS_PATH length, lowest ORIGIN type (IGP beats EGP beats Incomplete), lowest MED (only compared between routes from the same neighboring AS), eBGP preferred over iBGP, lowest IGP metric to the NEXT_HOP, and finally lowest BGP Identifier (Router ID) as a final tiebreaker. Cisco's implementation checks a proprietary Weight attribute before all of these, but Weight is not part of the RFC 4271 specification.