Reducing blocking probability and QoT violation in dynamic elastic optical networks via load-aware margin selection

In this paper, we study the effect of load-aware margin selection on the performance of elastic optical networks (EONs) in dynamic service provisioning. First, a mixed-integer linear programming (MILP) formulation is proposed as a benchmark for dynamic routing, modulation level, spectrum, and power assignment (RMSPA). In this formulation, nonlinear interferences (NLIs) in optical fibers and the noise of amplifiers are considered. In addition, a margin which is calculated based on the network load is considered to estimate the additional noise which may degrade the signal quality of transmission (QoT) during its life time. The objective of this formulation is to minimize the spectrum usage and impairments of/from the signal which is optimized by choosing the optimum channel among a number of pre-calculated channels based on the QoT constraints. In order to reduce the RMSPA run time, heuristic algorithms are also proposed. The algorithms employ K -least weighted paths for routing. The defined weights are distance, the amount of traffic load, and a mixture scheme. Furthermore, the algorithms try to minimize the spectrum usage by choosing the highest possible modulation level while considering a specific criterion. The assumed criteria are: i) choosing the lowest possible power spectral density (PSD) in a discrete set, ii) selecting the optimum power among the set, iii) obtaining maximum SNR. We simulated the proposed algorithms and the MILP, and compared their performances on NSFNET and US Backbone topologies. The results reveal that the constrained best SNR scheme has the best performance among the heuristics and its performance is close to MILP, especially in higher loads. Moreover, results show that considering the links congestion leads to blocking reduction. We also simulated the first-fit based algorithms without considering NLIs which assign the modulation based on: i) considering only amplifier noise and self-channel interference (SCI), ii) experimental distance thresholds, and iii) overestimating the NLI. The results imply that the proposed MILP and heuristic algorithms outperform the other algorithms in terms of the summation of outage (probability of QoT violation) and blocking probability.