Shared Protection-Based Virtual Network Embedding Over Elastic Optical Networks

Deep reinforcement learning for infrastructure as a service over flexible optical networks

“…DRL presents an opportunity to effectively search the solution space and learn a superior allocation policy. In the literature, multi-agent DRL approaches have been applied to VONE because the large combinatorial space of node-and pathselections has been considered too large for a single agent [9] . However, multi-agent DRL is more complex and may lead to less-optimised solutions due to the difficulty of agent cooperation.…”

“…However, multi-agent DRL is more complex and may lead to less-optimised solutions due to the difficulty of agent cooperation. Separate agents to allocate nodes and links [8] or nodes, links and backup resources [9] have been proposed and an arbitrary selection of heuristics used for performance comparison. The latter is a result of a lack of heuristic benchmarks in this area.…”