Chapter 6. BGP Cores and Network Scalability

Border Gateway Protocol (BGP) is the routing protocol that glues the Internet together. It falls under the Exterior Gateway Protocol (EGP) category—unlike the protocols described in previous chapters, which are Interior Gateway Protocols (IGPs). BGP version 4 is the current version, but throughout this book, it is referred to simply as BGP.

Traditionally, BGP has been used to exchange routing information between different autonomous systems. In the typical configuration, BGP ties Internet service providers (ISPs) to their customers and each other. This chapter does not deal with connections to the Internet or inter-ISP operations. Instead, it presents the proven, robust, and scalable BGP features that allow your network to grow past IGP limitations. The only portion where Internet connectivity is dealt with explicitly is in a series of case studies in the "External Connections" section.

This chapter is not about BGP itself, but how you can use it to scale your network even further. To understand all the concepts covered in this chapter, you need to be familiar with the basic operation of BGP. If you need a quick review, read Appendix D, "Border Gateway Protocol 4 Basics of Operation," before continuing.

As described in Chapters 1 and 2, hierarchy, addressing,summarization, and redundancy are essential components of good network design. Chapters 3, 4, and 5 took this one step further, and described how the techniques used in deploying an IGP is also important. However, there are limits in all network designs, and as your network grows, you will inevitably hit them. The main limitations are the sheer volume of routing information carried in the routing protocol, and the increasing levels of policy required to successfully manage the network. On the other hand, BGP is currently deployed worldwide and carries approximately 155,000 routing entries at the core of the Internet. (This number is growing at the time of this writing. You can track the growth of the Internet by visiting the "CIDR Report" website at http://www.cidr-report.org/.) Some providers have been known to carry closer to 280,000 routes.

Policies are hard to define and enforce with an IGP because of the limited flexibility. Usually, a tag is the only tool available.

In the age of increasingly complex networks (in both architecture and services), BGP offers an extensive suite of knobs to deal with complex policies, such as the following (to name just a few):

• Communities

• AS_PATH filters

• Local preference

• Multiple exit discriminator (MED)

BGP also counters instability by implementing a route dampening algorithm. This algorithm is applied when the advertisement of a route is suppressed if it is known to change regularly over time. (All the parameters from the periodicity of the flaps to the type of routes suppressed are configurable.) Although you will follow the structural recommendations given in this book when building networks with the different IGPs studied, BGP is not tied to a set hierarchical model. In fact, the topology can take any form, and the protocol will adapt to it. Look at the Internet, which has no discernible hierarchical structure; it is impossible to pinpoint a core or a distribution layer for the Internet as a whole—and it works.

Neighbors, Routes, and Propagation Rules in BGP A router using BGP exchanges routing information by forming a neighbor relationship with other routers. BGP routers can establish either internal or external peerings. BGP peers in the same autonomous system (AS) are called iBGP peers, whereas peers in a different AS are called eBGP peers. An AS might have more than one external connection; in that case, it is necessary to have several BGP speakers in the network to maintain routing consistency. Unlike other protocols, the rules of when and if a prefix is advertised to a neighbor depend on the type of neighbor the prefix was learned from. Also, a route is propagated only if it has been selected as the best for a specific destination. (For more details, read Appendix D.) Three combinations are possible: Routes learned from an eBGP peer?/span> Propagated to all peers Routes learned from an iBGP peer?/span> Propagated only to eBGP peers Routes originated locally?/span> Propagated to all peers Because routes learned from iBGP peers are not sent to other iBGP peers, it is clear that a full logical mesh is needed between them to ensure consistent routing information.

This chapter is a discussion of the use of BGP as a way to scale your network even further. The discussion starts with a description of the implementation in the core of the network (where full routing is required) and expands the concepts to be used in the network as a whole.