On Retrieval Order of Statistics Information from OpenFlow Switches to Locate Lossy Links by Network Tomographic Refinement

To maintain service quality and availability in managed networks, detecting and locating high loss-rate links (i.e., lossy links that are likely congested or physically unstable) in a fast and light-weight manner is required. In our previous study, we proposed a framework of network-assisted location of lossy links on OpenFlow networks. In the framework, a measurement host launches a series of multicast probe packets traversing all full-duplex links; and then the controller retrieves statistics on the arrival of those probe packets at different input ports on different switches and compares them to locate high loss-rate links. The number of accesses to switches required to locate all lossy links strongly depends on the retrieval order in collecting the statistics and should be small as much as possible. Therefore, in this paper, to minimize the necessary number of accesses, we develop a new location scheme with an appropriate retrieval order using a Bayesian-based network tomography to refine candidates for lossy links. The results of numerical simulation on a real-world topology demonstrate the effectiveness of the new location scheme.11th International Conference on Intelligent Networking and Collaborative Systems(INCoS 2019), September 5-7, 2019, Oita, Japa