Throughput and latency targeted RL spectrum allocation in heterogeneous OTN

Abstract

The increased adoption and development of 5 G-based services have greatly increased the dynamic nature of traffic, including types, sizes, and requirements. Flex-grid elastic optical networks (EONs) have become prolific in supporting these services. However, this transition led to issues such as lower traffic throughput and resource wastage in the form of bandwidth fragmentation. With the continued growth of these services, proper traffic management to reduce this issue has become essential. To overcome this challenge, we propose a Throughput and Latency-First Reinforcement Learning-based spectrum allocation algorithm (TLFRL) in IP-over-fixed/flex-grid optical networks. The main target of TLFRL is to reduce the need to reallocate the spectrum by lowering the fragmentation and blocking probability. We achieve this by leveraging advanced demand organization techniques while using traditional networking infrastructure intelligently to offload compatible services, avoiding latency violations. Extensive simulations evaluated traffic throughput, fragmentation, and average latency. The results show that the proposed solution outperforms contemporary fixed grid-based and heuristic approaches. It also provides comparable results to state-of-the-art flex-grid spectrum allocation techniques.