Cross-Layer Resource Provisioning for Long-Haul Fiber Optic Networks

Elastic optical networks (EONs) have been proposed to meet future communication demands. Planning the resource usage of EON has been the subject of extensive research. Routing and spectrum assignment (RSA) algorithms are used to minimize network resources used. Scalability, non-optimality, and computation complexity remain a problem for heuristic algorithms in the published literature. In addition, estimating of physical-layer impairments (PLIs) in EON is important in the network planning stage for long-haul systems. The widely applied transmission reach (TR) model ignores the network state, and is therefore too conservative, leading to resource over-provisioning. The so-called Gaussian noise (GN) model is one of the most accurate PLI estimates. However, the GN model has limited compatibility because this nonlinear model is complex and designed for deterministic traffic.

In this dissertation, we comprehensively study the resource provisioning problem from three different perspectives: (A) more scalable offline mixed integer linear programming (MILP) algorithms with linearized PLI models; (B) a novel analytic PLI estimate for random bandwidth traffic; (C) probabilistic online-offline combined resource provisioning algorithms with realistic traffic.

In (A), we propose a link based recursive MILP (re-MILP) algorithm that balances the scalability and optimality. A linear physical-layer estimation model based on the GN model, referred to as the conservative linearized Gaussian noise (CLGN) model, is also proposed. The re-MILP algorithm using the CLGN saves significant network resources compared to the benchmark algorithm with the TR model.

In (B), we propose a GN-model-based PLI estimation for random bandwidth demands, referred to as the probabilistic spectrum GN (PSGN) model. The proposed PSGN model has a simple form and controllable reliability, and it outperforms the GN model in estimation accuracy for random bandwidth demands. The PSGN model is also suitable for estimating PLIs of arbitrary pulse-shaped demands.

In (C), we propose the flexible-online-offline probabilistic (FOOP) algorithm for aggressively allocating spectrum for realistic time-varying demands. State-of-the-art provisioning, called standard provisioning, reserves the maximum resources needed by the time-varying demands. The FOOP algorithm models the network resources in a probabilistic way. We also optimize the use of regeneration resources in the network using the PSGN model. Compared to standard provisioning, the FOOP algorithm saves significant spectrum resources. Moreover, as shown through simulation, the PSGN model saves regeneration resources in a continental-scale network compared to the TR model.

For each part of the research, we evaluate the proposed algorithms and compare their performance with published algorithms as benchmarks. The research in this dissertation is a complete and comprehensive study of provisioning for long-haul EONs, and the proposed algorithms are scalable, near-optimal, and computationally feasible.