BGP eBGP vs iBGP

Introduction

Border Gateway Protocol (BGP), is a protocol used for exchanging routes on a grand scale. It's old (latest version from 1995), slow (converging takes forever) and extremely scalable (laughs at areas/levels of IGPs). Because of its modularity it is also very flexible.

It's a path-vector type routing protocol, where traversed Autonomous Systems (AS) are the path and the interface from which to send a packet is the vector.

One of the most important characteristics of BGP is its scalability due to rarity of advertisements between neighbors. Whole routing knowledge is exchanged only during the session initiation. After that routes are advertised on "need to know" basis - i.e. when new information is learned or old information withdrawn.

BGP doesn't have its own connectivity protocol for communication between peers. It uses TCP port 179 to achieve reliable connections. Because of that, it's often argued that BGP isn't really a protocol per se, it's more of an application running on TCP protocol (like Telnet).

BGP is defined by:

• RFC 1771, "A Border Gateway Protocol (BGP-4)"
• RFC 1772, "Application of the Border Gateway Protocol in the Internet"
• RFC 1966, "BGP Route Reflection: An Alternative to Full-Mesh I-BGP"
• RFC 1997, "BGP Communities Attribute"
• RFC 2270, "Using a Dedicated AS for Sites Homed to a Single Provider"
• RFC 2385, "Protection of BGP Sessions through the TCP MD5 Signature Option"
• RFC 2439, "BGP Route Flap Damping"
• RFC 2842, "Capabilities Advertisement with BGP-4"
• RFC 2858, "Multiprotocol Extensions for BGP-4"
• RFC 2918, "Route Refresh Capability for BGP-4"
• RFC 3065, "AS Confederations for BGP"

Neighbors

EBGP vs IBGP

BGP forms two types of relationships - external (EBGP) and internal (IBGP).

External relationships are by default limited to immediate physical connections by TTL equal 1. EBGP sessions are therefor set up using neighbor facing interface IP address. Default TTL can be increased with multihop command, which allows to establish remote relationships or use loopbacks to maintain a single session over multiple redundant links.

EBGP has one simple advertising rule - forward all active BGP learned routes to all BGP neighbors.

Internal relationships are usually setup over multiple hops - default TTL is 64 (same value as multihop external sessions and confederations). Suggested method of setting up session is using loopbacks to allow IGP to reroute in case of interface failure, which aids stability.

Because IBGP doesn't have a mechanism for loop detection (like AS Path attribute in EBGP) advertising rules were modified to avoid forming routing loops - forward all active EBGP learned routes to all BGP neighbors, however forward active IBGP learned routes only to EBGP neighbors.

Due to that rule, maintaining full mesh connectivity between internal peers is a common practice.