Vasantha Kumar Hosahalli Seenappa
Hyperscale MPLS/IP backbone networks require high levels of availability and operational reliability to support the interconnection of distributed Points of Presence across geographic regions. Despite this requirement, most existing frameworks for automating network management execute configuration changes based on a "fire and forget" principle with reactive failure recovery mechanisms that will often lead to service disruptions due to lengthy rollback time frames post-failure. The research proposes a framework for the autonomous management of hyperscale MPLS/IP backbone network called the Predictive Configuration Governance (PCG) framework. The focus of the research is to proactively predict deployment risk based on operator intent, OpenConfig/NETCONF (Network Configuration Protocol) configurations, topology, telemetry data, configurations change and previous incidents history. A Configuration Risk Score will allow automated deployment of low-risk changes while routing high-risk configurations to be manually approved. In addition, the PCG framework will continuously refine the model prediction accuracy via a refined closed-loop feedback process as post-deployment telemetry continues to be ingested. The conceptual framework we present enhances prediction intelligence with policy guardrails that can be applied to automated deployments in a core network environment to ensure that the deployment is done safely. Conceptual evaluations have shown that the proposed framework has the capacity to greatly improve the success rates of configurations, reduce the mean time to recover and lower the rollback rate when compared to traditional, non-governed automation pipelines in a production system. Ultimately, the research presented here offers a scalable way to improve the stability and robustness of next-generation IP backbones.