Deep reinforcement learning-aided fragmentation-aware RMSA path computation engine for open disaggregated transport networks

Optical network control platforms are facing an unprecedented increase of complexity due to the requirements for openness and optical system dis-aggregations. While Reinforcement Learning (RL) and Deep RL (DRL) methods can be used for network control decision-configuration solutions, these methods can be unsuitable to find correct policies to control open disaggregated transport networks (ODTN). In this context, this article proposes DeepSF-PCE, a single-agent deep reinforcement learning spectrum fragmentation aware path computation engine to solve the Routing, Modulation and Spectrum Assignment (RMSA) problems in ODTN. DeepSF-PCE engine learns fragmentation-aware policies that maximize the number of allocated service requests. Simulation results show that DeepSFPCE can increase by more than 40% the number of configured optical connectivity services compared to existing solutions.