Master-Slave Agent Architecture with Deep Reinforcement Learning for Efficient Spectrum Allocation in Multicore Elastic Optical Networks

Presenter(s)

Subbulakshmi Easwaran

Abstract or Description

Elastic Optical Networks (EONs) represent a groundbreaking advancement in flexible network architecture, enabling the transmission of diverse data signals through optical fiber channels. By allowing dynamic resource management, EONs can optimize bandwidth allocation, ensuring that data transmission is efficient and reliable. Multicore Elastic Optical Networks (MC-EONs) represent a groundbreaking solution to the escalating needs for ultra-high bandwidth and energy efficiency in today’s communication networks. Yet, tackling the Routing, Modulation, and Spectrum Assignment (RMSA) problem in MC-EONs presents significant challenges due to the intricate interdependencies among cores and the necessity for dynamic spectrum management. This paper introduces an innovative Master-Slave Agent architecture driven by Deep Reinforcement Learning (DRL) to effectively overcome these hurdles. The Master agent oversees holistic decision-making by synthesizing state information from across the network. In contrast, numerous Slave agents function locally within each core, optimizing resource allocation and responding adaptively to varying traffic conditions. To further improve scalability and flexibility, we propose a Cloneable Slave mechanism, allowing for the swift deployment of pre-trained agents in new cores and situations. Simulation outcomes indicate that our framework significantly surpasses conventional methods regarding spectrum efficiency and resilience to fluctuating traffic patterns. This research underscores the Master-Slave Agent paradigm's potential in deftly managing the complexities inherent in MC-EONs.