Network Topology-Traceable Fault Recovery Framework with Reinforcement Learning

Abstract

Network Function Virtualization (NFV) is the focus of much attention especially in the field of network operation automation because of its efficient usage of resources and lower capital expenditures. However, NFV introduces additional complexity into network monitoring and management due to the network function being independent of the hardware. Moreover, maintaining the workflow still requires tremendous effort in order to sustain the automation function. In fact, a vast amount of workflows or program codes for automated deployment and the failure recovery operation have to be created and maintained per type of service and failure case. To address the above problems, previously we proposed an artificial intelligence-assisted workflow management framework. This work demonstrated that a scheme for fault-recovery operation automation. However, the conventional issue is to fix the dimensions of the input vector for machine learning (ML) algorithms to prevent the relearning repetition even if the network configuration and topology is changed. To address the above issue, this paper proposes a reinforcement learning-based fault recovery framework by applying the reinforcement learning (RL) algorithm which can adapt to changes of network topology and configuration. We demonstrate the effectiveness of the proposed framework with testbed results.