Centralized routing decisions provide enormous flexibility that often comes at the cost of the robustness provided by distributed protocols. While centralized control, such as Software Defined Networking (SDN), enables easy configuration and has high manageability, distributed routing protocols tend to have faster path installation and failure response rates. In their 2015 paper, “Central Control Over Distributed Routing”, Vissicchio, Tilmans, Vanbever, and Rexford introduce Fibbing, an architecture that achieves both flexibility and robustness through central control over distributed routing. Fibbing introduces fake nodes and links into an underlying link-state routing protocol, so that routers compute their own forwarding table based on the augmented topology.
Here, we aim to reproduce a subset of the results presented in their paper, namely the time it takes for a certain percentage of nodes to change next-hops (Figure 8a); number of fake nodes created for a certain percentage of nodes to change next-hops (Figure 8b); throughput evolution of Case Study presented (Figure 9); throughput evolution in relation to failure responses (Figure 10). To reproduce the Figures 8a and 8b, we generate random forwarding requirement DAGs, and apply the author’s implementation of the algorithms within a VM. For reproducing the case studies (Figure 9 and Figure 10), we set up a testbed in mininet.
While we’re able to reproduce certain aspects of Figures 8a, and 8b as done in the paper from Simple, we ran into additional friction on Merger, and our results differ from the paper. For Merger, our reproduction has a constant augmentation of fake nodes, while the runtime of the algorithm seems to correlate with the % of nodes changing their next hop.
We are able to replicate the results of the Case Study in Figure 9b and 10. While the trend are demonstrated to be largely similar, we ran into small discrepancies that are believed to arise from the way we collected our data.