On robust network coding subgraph construction under uncertainty

We consider the problem of network coding subgraph construction in networks where there is uncertainty about link loss rates. For a given set of scenarios specified by an uncertainty set of link loss rates, we provide a robust optimization-based formulation to construct a single subgraph that would work relatively well across all scenarios. We show that this problem is coNP-hard in general for both objectives: minimizing cost of subgraph construction and maximizing throughput given a cost constraint. To solve the problem tractably, we approximate the problem by introducing path constraints, which results in polynomial time-solvable solution in terms of the problem size. The simulation results show that the robust optimization solution is better and more stable than the deterministic solution in terms of worst-case performance. From these results, we compare the tractability of robust network design problems with different uncertain network components and different problem formulations.