Parallel Differential Evolution Meta-Heuristics and Modelling for Network Slicing in 5G Scenarios
Résumé
Network Slicing is the crucial component to face the Internet’s ossification and support the heterogeneity of 5G scenarios. A slice is a virtual network mapped over a real network. Mathematically, the mapping is known as Virtual Network (VN) Embedding (VNE), which is the mapping of virtual nodes and links to real nodes and links obeying the QoS parameters present in the VN request and available resources. Since this is an optimization and NP-hard problem, multiple efforts have been made to create VNE algorithms. Considering such efforts, this work presents:(i) a fitness function regarding multiobjective optimization to maximize the quantity of embedding; (ii) two new parallel Differential Evolution (DE) approaches to face the VNE problem; and (iii) a Stochastic Petri Net to model our approaches. We designed these two versions due to the lack of viable parallel solutions in the 5G scenario. We compared three approaches with two different infrastructure datasets and nine DE setups. The results demonstrate that the better parallel version reduced the runtime by 68.94% and 51.03% using datasets 1 and 2, respectively, and kept the same acceptance rate. The parallel performance decreases the runtime in certain conditions, and we explain in which scenarios the parallel approaches obtain advantages and disadvantages.