An Approach based on Reinforcement Learning for Management of Abnormal Communications via Fixing Faulty Components in Software-Defined Networking

In recent years Software-Defined Networking (SDN) has been proposed as a new solution for computer networks. Computer networks based on SDN architecture may be annoyed by plenty of various software or hardware faults that lead to abnormal communications. Hence, immediate detection and localization of faults are indispensable to preserve the safety of the network and reliable communications. In fact, localization of faulty components is the main phase of fault management in SDN. In this research, the problem of fault localization in SDN is studied and proposed an approach based on Reinforcement Learning (RL) with properly selected features such as cost tests, network traffic, etc. The main aim of the fault localization approach is to have a reliable communication among various network components. We measured the operation and efficiency of the proposed approach by vast simulations. The proposed approach can localize faulty devices at a lower cost in the network. Results of simulations demonstrate that our proposed approach is better than the other fault localization approaches in SDN.

کلیدواژه ها:

Computer networks ، fault management ، passive evaluations ، reinforcement learning ، reliable communication.