Practical Programmable Topologies in the WAN

With the rapid adoption of Reconfigurable Optical Add-Drop Multiplexers (ROADMs) in wide-area networks (WANs), backbone providers have the ability to change the physical layer of their networks by programming wavelengths on each edge. Yet the physical layer of today’s WANs has remained static because (i) there is no simple way to predict the throughput gains, and (ii) there is no clear path to the practical insertion of programmable topologies into the already complex traffic engineering (TE) schemes. To overcome these challenges, we first provide bounds on throughput gains of programmable WANs, propose a simple heuristic to estimate gains by considering the physical topology, and present opportunities for programmability by studying traffic traces. Second, we argue that the path to adoption should not involve rewriting cross-layer TE formulations. Instead, we propose a Reflow graph as an abstraction that permits topology programmability in the TE algorithms. A Reflow graph augments the physical graph and uses it as input to the unmodified TE algorithm, capturing a variety of TE formulations. As such, we can also directly transfer standard consistent network update schemes to optimize across layers. Moreover, we evaluate our abstraction using a testbed and simulations: our results show that Reflow has competitive computation times and is feasible in practice, allowing to quickly react to unplanned events such as fiber cuts.