Dynamic RAN Slicing Mapping Under the Sliding Scheduled Traffic Model Based on Deep Reinforcement Learning in Elastic Optical Networks

To relieve network resource pressure and avoid network congestion, the adoption of a sliding scheduling traffic model is crucial to achieve flexible radio access network (RAN) function deployment and provide customized services. In this paper, we investigate the problem of dynamic RAN slicing mapping under the sliding scheduled traffic model in metro-access/aggregation elastic optical networks (EONs). These particular sliding scheduled RAN slicing requests allow for an initial latency during the setup phase, as long as resource allocation is available prior to a preset deadline. To combine RAN slicing request scheduling in the time domain with routing and spectrum assignment (RSA) in the spectrum domain, we propose a dynamic RAN slicing mapping (DRSM) algorithm for EONs based on deep reinforcement learning (DRL). In our DRSM method, we introduce a time-frozen scheme aimed at reducing the action space to accelerate the training process. Simulation results on a 30-node network topology demonstrate that the proposed DRL-based DRSM method outperforms the two baseline heuristics, potentially saving at least 48.1% more blocking probability under the sliding scheduled traffic model with different arrival rates and sliding factors.