This document is the result of the final degree project, whose objective is to develop a
diagnosis fault system based on bayesian networks for Software Defined Networks (SDN).
This system provides, provoking some situations in the simulated network, a set of
possible faults and their probability of happening. This is possible thanks to different
technologies such as OpenFlow, Mininet, POX, or SMILE, that allows to simulate a network and take data in order to process and obtain by means of a bayesian network the diagnosis result.
To carry this out, the topology has been defined, with a set of networking rules and
system resources, and a streaming service has been simulated, in which we have generated the faults. After that, the network data have been collected thanks to the controller, and processed by a monitoring module, during the different generated situations.
The collected data of the simulated network have been used to feed a learning algorithm, allowing in this way to generate a diagnosis model in a bayesian network, that after being integrated in the system, it offers the most likely root cause failure of the possible range.
Then, three diagnosis models have been generated in order to evaluate them in the
scenario in question. Finally, some possible future lines of work to improve the diagnosis system are presented.