Towards An Adaptive Selection Of Loss Estimation Techniques In Software-Defined Networks

Next generation Software-defined Networks (SDN) aim at deeply programmable switches which can be leveraged by SDN controllers to offload self-contained, logically persistent tasks. One such task is flow and network monitoring, specifically, fault detection and loss estimation, which is essential for SDN applications that provide quality of service guarantees under network dynamics. In this work, we devise, implement, and evaluate fault detection and loss estimation techniques built upon tasks devolved to SDN switches. Subsequently, we contribute (i) an analysis and empirical evaluation of the benefits and costs of different packet loss estimators depending on the network conditions; and (ii) a case study showing how an adaptive monitoring framework, which flexibly exchanges the estimation techniques, would retain a thoroughly good fidelity while optimizing the monitoring costs.