This paper proposes DeepRMSA, a deep reinforcement learning framework for routing, modulation and spectrum assignment (RMSA) in elastic optical networks (EONs). DeepRMSA learns the correct online RMSA policies by parameterizing the policies with deep neural networks (DNNs) that can sense complex EON states. The DNNs are trained with experiences of dynamic lightpath provisioning. We first modify the asynchronous advantage actor-critic algorithm and present an episode-based training mechanism for DeepRMSA, namely, DeepRMSA-EP. DeepRMSA-EP divides the dynamic provisioning process into multiple episodes (each containing the servicing of a fixed number of lightpath requests) and performs training by the end of each episode. The optimization target of DeepRMSA-EP at each step of servicing a request is to maximize the cumulative reward within the rest of the episode. Thus, we obviate the need for estimating the rewards related to unknown future states. To overcome the instability issue in the training of DeepRMSA-EP due to the oscillations of cumulative rewards, we further propose a window-based flexible training mechanism, i.e., DeepRMSA-FLX. DeepRMSA-FLX attempts to smooth out the oscillations by defining the optimization scope at each step as a sliding window, and ensuring that the cumulative rewards always include rewards from a fixed number of requests. Evaluations with the two sample topologies show that DeepRMSA-FLX can effectively stabilize the training while achieving blocking probability reductions of more than 20.3% and 14.3%, when compared with the baselines.

The combination of elastic optical networks (EONs) and software-defined networking (SDN) leads to SD-EONs, which bring a new opportunity for enhancing programmability and flexibility of optical networks with more freedom for network operators to customize their infrastructure dynamically. In this paper, we investigate how to apply multidomain scenarios to SD-EONs. We design the functionalities in the control plane to facilitate multidomain tasks, and propose an interdomain protocol to enable OpenFlow controllers in different SD-EON domains to operate cooperatively for multidomain routing and spectrum assignment. The proposed system is implemented and experimentally demonstrated in a multinational SD-EON control plane testbed that consists of two geographically distributed domains located in China and USA, respectively. Experimental results indicate that the proposed system performs well for resource allocation across multiple SD-EON domains.

This paper proposes a self-taught anomaly detection framework for optical networks. The proposed framework makes use of a hybrid unsupervised and supervised machine learning scheme. First, it employs an unsupervised data clustering module (DCM) to analyze the patterns of monitoring data. The DCM enables a self-learning capability that eliminates the requirement of prior knowledge of abnormal network behaviors and therefore can potentially detect unforeseen anomalies. Second, we introduce a self-taught mechanism that transfers the patterns learned by the DCM to a supervised data regression and classification module (DRCM). The DRCM, whose complexity is mainly related to the scale of the applied supervised learning model, can potentially facilitate more scalable and time-efficient online anomaly detection by avoiding excessively traversing the original dataset. We designed the DCM and DRCM based on the density-based clustering algorithm, and the deep neural network structure, respectively. Evaluations with experimental data from two use cases (i.e., single-point detection and end-to-end detection) demonstrate that up to 99% anomaly detection accuracy can be achieved with a false positive rate below 1%.

High spectrum efficiency and fast restoration speed are highly desired for survivable elastic optical networks (EONs). In this paper, we take the advantages of failure-independent path-protecting preconfigured cycles (FIPP p-cycles) and investigate how to realize spectrum efficient resilience design with them. We first study the problem of offline service provisioning with FIPP p-cycles. We formulate an integer linear programming model and prove that the problem is NP hard. Then, several time-efficient heuristics are designed for FIPP p-cycle formulation and related routing, modulation format, and spectrum assignment. Extensive simulations on offline provisioning verify that the heuristics can obtain near-optimal solutions. Next, we consider online service provisioning with FIPP p-cycles in dynamic EONs. In order to overcome the decrease of protection efficiency during dynamic network operation, we propose a p-cycle reconfiguration scheme to reoptimize protection structures on-the-fly. Simulation results demonstrate that the proposed algorithms can improve spectrum efficiency and reduce bandwidth blocking probability simultaneously.

Availability-aware service provisioning tries to satisfy the availability requirements of clients on demand and can improve the resource efficiency in optical networks. This paper studies the problem of flexible availability-aware differentiated protection (ADP) in elastic optical networks (EONs). We first consider the unique features in EONs, e.g., flexible spectrum allocation and bandwidth-squeezed protection, describe the problem of ADP, and present a theoretical analysis on service availability. Then, we develop an ADP algorithm that can change the path protection scheme to adapt to different service availability requirements. An availability-aware backup reprovisioning strategy is also proposed as a supplement of the ADP algorithm to further improve the protection efficiency. We also discuss the system design for OpenFlow-based software-defined EON (SD-EON) framework to facilitate the ADP scheme. Finally, to demonstrate the overall design, we construct an SD-EON control plane testbed that consists of 14 standalone nodes, implement the proposed algorithms in the OpenFlow controller, and perform experiments to evaluate their performance. The results indicate that the system performs well, and the proposed algorithms can reduce blocking probability effectively without sacrificing the service availability of requests.

We investigate dynamic pre-configured-cycle (p-cycle) protection design for spectrum-sliced elastic optical networks (EONs). Several novel algorithms are proposed to use the spectrum planning of working and backup resources. We first combine the protected working capacity envelope (PWCE) p-cycle design with spectrum planning, and design an algorithm that can achieve dynamic p-cycle design in EONs. Then, to resolve the coverage issue of the PWCE-p-cycles, we consider dynamic p -cycle design with Hamiltonian cycles and propose to use topology partition for reducing the lengths of backup paths. Simulation results show that our proposed algorithms achieve lower blocking probability than the shared path-protection (SPP) algorithm, while the average length of backup paths per request can be controlled well. To the best of our knowledge, this is the first attempt to address dynamic p-cycle protection design with spectrum planning for EONs.