| Abstract: | The servitization of network resources leads to new challenges for optical networks. For instance, to provide on-demand lightpaths as a service while keeping the probability of packet loss (PPL) low, issues such as lightpath setting up, resource reservation and load balancing must be addressed. We present a self-adaptive framework to process lightpath requests on packet switching optical networks that considers and handles the aforementioned issues. The framework is composed of a dimensioning phase that adds up new resources to an initial topology and a learning phase based on reinforcement learning that provides self-adaptation to tolerate traffic changes. The framework is tested on three realistic mesh topologies achieving a PPL between (1 times 10^{-1}) and (1 times 10^{-6}) for different traffic loads. Compared to fixed multi-path routing strategies, our framework reduces PPL between (19%) and up to (80%). Furthermore, no packet loss can also be achieved for traffic loads equal to or lower than 0.4. |
