Virtualized controller placement for multi-domain optical transport networks using machine learning

Abstract

Optical multi-domain transport networks are often controlled by a hierarchical distributed architecture of controllers. Optimal placement of these controllers is very important for efficient management and control. Traditional SDN controller placement methods focus mostly on controller placement in datacenter networks. But the problem of virtualized controller placement for multi-domain transport networks needs to be solved in the context of geographically distributed heterogeneous multi-domain networks. In this context, edge datacenters have enabled network operators to place virtualized controller instances closer to users, besides providing more candidate locations for controller placement. In this study, we propose a dynamic controller placement method for optical transport networks that considers the heterogeneity of optical controllers, resource limitations at edge hosting locations, and latency requirements. We also propose a machine-learning framework that helps the controller placement algorithm with proactive prediction (instead of traditional reactive threshold-based approach). Simulation studies, considering practical scenarios and temporal variation of load, show significant cost savings compared to traditional placement approaches.