Experimental Observations and Analysis of BGP Route Flapping Dynamics

The Border Gateway Protocol (BGP) is responsible for updating inter-domain routing information. One of the poorly understood aspects of BGP behaviour that has a direct impact on Internet stability is route flapping, where a route is repeatedly advertised but then withdrawn. This behaviour leads to slow convergence times resulting in Autonomous Systems (ASes) having inconsistent views of the Internet. Route flapping is an important matter that needs to be addressed to ensure the stability of the Internet as it continues to grow. However, there is currently no satisfactory solution to prevent or mitigate the effects of route flapping. Before a satisfactory solution can be developed, a better understanding of how route flapping originates and propagates is needed. In this paper, we used our large network testbed to construct a network with twenty ASes. We were able to generate route flapping and demonstrated that it can be caused by a congested link between ASes. We demonstrated that the frequency of route flapping is proportional to the level of congestion. We also showed that route flaps propagate across the entire network. As part of this research, we developed an automation tool to construct medium-to-large BGP testbeds where all interconnected ASes are placed under a common out-of-band management domain, providing a single vantage point to observe the behaviour and propagation of routing flaps across affected BGP peers. As far as we are aware this is the first time route flapping has been generated and analysed using a large scale testbed of commercial equipment. However, others have carried out research using simulation and modelling. We found our results consistent with theirs.