A Distributed, Parallel and Fault Tolerant BGP Routing Daemon

Abstract

The protocol that has delivered the Internet to the world suffers from aged implementations in current core routers. The implementations lack proper fault tolerance and suffer from single-threaded processing, underutilizing multicore processors. While a lack of multi-core processing capabilities was not a notable deficit 20 years ago, today a modular router is a small compute cluster in itself with multiple multi-core processors. This thesis identifies first that in order to simplify, scale and improve fault tolerance in a BGP implementation, it can be decomposed into smaller parts. This thesis also shows that the processing of a BGP implementation’s critical path can be parallelized down to a per-prefix level and distributed over multiple nodes in a cluster implementation of a distributed RIB. Using the ErlBGP architecture, greater stability could be achieved in the Internet.