Border Gateway Protocol – Complexity not Scale
A common misconception that Border Gateway Protocol ( BGP ) is used solely for network scalability, replacing Interior Gateway Protocol ( IGP ) once a specific prefix or router count has been reached.
Although BGP does form the base for large networks, a properly designed IGP can scale tens of thousands of routers. BGP is not just used for scalability, it is used to decrease the complexity of networking, rather than size. The key to efficient routing protocol design is to start with business design principles and break failure domains into small pieces. Keeping things simple with BGP is key to stabilizing large networks.
What usually starts out as a single network, easily turns into multiple networks as the business grows. Easier to split networks into small pieces and to “aggregate” the information as much as possible. Aggregating routing information hides parts of the network, but it also speeds up convergence on link / node failure.
Policy-oriented control plane reduces network complexity
BGP is a policy-oriented control plane-routing protocol and is used to create islands of networks that match business requirement to administrative domains. When multiple business units present unique requirements, it is hard to design all those unique requirements using a single set of routing policies. BGP can be used to decrease policy complexity and divide the complexity into a manageable aggregation of policies.
Two business units, for example, HR, represented by a router on left and Sales department, represented by a router on the right. The middle networks form a private WAN, used simply as transit. Business has decided that both these networks should be treated differently and have different traffic paths. HR must pass through the top section of routers and Sales must pass through the bottom half of routers. With an Interior Gateway Protocol ( IGP ), such as OSPF, traffic engineering can be accomplished by manipulating the cost of the links to influence traffic path.
However, the metrics on the links must be managed on a per destination basis. If you have to configure individual links per destination; it would become almost impossible to do with a link-state IGP. If BGP is used, this logic can be encoded using either Local Preference or Multiple Exit Discriminator. Local preference is used for single AS design and MED used for multiple AS. Local preference is local and does not traverse multiple AS.
Networks grow and should be allowed to grow organically. Each business unit may require a number of different topologies and design patterns. Trying to design to all these different requirements would increase network complexity. In the context of a single IGP, it may add too many layers of complexity. BGP provides a manageable approach to policy abstraction by controlling specific network traffic patterns within and between Autonomous Systems.