Sparsifying Network Topologies for Application Guidance

Topology managers expose network topology information to applications to improve application-level resource management. An example for various ongoing standardization activities is Application-Layer Traffic Optimization (ALTO). Due to privacy constraints, an exposed network topology has to be abstracted by a policy system. This can be achieved by graph sparsification algorithms. This paper analyzes different graph clustering algorithms for a topology manager offering a policy-based abstract view on the network. We both propose solutions without and with additional labeling indicating relevant destinations for the corresponding application. Those algorithms are integrated into an import system leveraging existing route policies. Given the complexity of the problem, our algorithms are heuristics. We quantify the abstraction error and compare the trade-off between topology size and accuracy. Our comparison reveals that clustering algorithms typically outperform graph sparsification algorithms. We also show that even without label information in many cases one can significantly reduce the size of a topology map without dramatically impacting application guidance accuracy.