By Mark Goodwin, Cox Communications, Inc.
The Border Gateway Protocol (BGP) is a critical routing protocol used in large-scale networks, including the Internet. From the viewpoint of an organization’s networking, BGP serves both internal and external purposes. External BGP (EBGP) is a prevalent method used for establishing connections to networks outside of your organization, serving the external use case. Conversely, for internal use, the Internal BGP (IBGP) is commonly adopted. This document will concentrate on the modern approach for scaling IBGP networks from the perspective of the architecture, the platform, and the routing.
The traditional method for building IBGP networks incorporates the IBGP full mesh model. This model is an excellent strategy for networks ranging from small to medium size. In this framework, all IP routing data is disseminated among all devices within the IBGP network. Essentially, each device maintains a logical connection or IBGP session with every other device. The total connectivity for the IBGP model is calculated using the formula 𝑁𝑁(𝑁𝑁−1)2 which is referred to as the n squared formula. Deployments with the IBGP full mesh model offer numerous benefits.
As the IBGP network expands, however, it faces several challenges. A fully distributed or regionalized route reflection model for IBGP is proposed to address these challenges, offering enhanced efficiencies and scalability for the IBGP network. This paper will outline the challenges associated with large-scaleIBGP and illustrate how the route reflection model effectively overcomes these challenges.
The traditional platforms for IBGP network consist of routing devices that serve as the network nodes.
These nodes boast considerable computational power, but this capacity must be distributed among various processes within the routing device. As the number of network devices escalates and the network’s complexity intensifies, it imposes a significant burden on the computational power, especially the processes designated for BGP processing. This paper proposes the use of x86 servers to handle the heavy workload of BGP processing. X86 based servers have significantly more computational power than the router platforms and offer substantial computational resources for BGP scalability. This paper will depict the x86 based server as a streamlined approach for achieving high BGP scalability.
In the IBGP full mesh network, routing is optimized as each node possesses a comprehensive map of the IBGP network and employs the shortest path computation from the IGP for efficient routing through the IBGP network. However, updates or modifications to the IBGP full mesh architecture can often lead to suboptimal routing. This paper will delve into some of the corner cases where this may occur and propose relevant solutions to preserve optimal routing.
This paper explores a modern approach to scaling IBGP networks, focusing on design principles and optimization for platform, architecture, and routing. We delve into the architectural considerations and propose enhancements to improve scalability and create efficiencies.