Evolutionary Algorithm with Phenotype Diversity for Virtual Network Embedding
Tatsuya Otoshi, Masayuki Murata · 2023
With the diversification of applications using the Internet, network virtualization technologies that flexibly allocate network resources are attracting attention. In network virtualization, virtual network embedding is important to properly map the requirements of the virtual network to the physical network. However, it takes time to calculate the solution by optimization, and recalculation of the embedding becomes a problem when the environment of the virtual and physical networks changes. Therefore, a method of having multiple solution candidates in advance and switching the solution depending on the situation is considered, but the design and updating of the solution candidates themselves remain an issue. Such a relationship between solution candidates and solution selection is similar to the relationship between genotype and phenotype in biological evolution, and it is a shortcut to get hints from evolution. In biological evolution, the phenotype searches for short-term practical solutions while the genotype continues to search for optimal solutions. By introducing this mechanism into the network, it is possible to select a quasi-optimal solution in a fluctuating environment while continuing the search for a better solution candidate itself. In this paper, we propose a dynamic virtual network embedding method in which the solution candidates themselves can be dynamically updated based on the evolution of genotype and phenotype. In this method, candidate solutions are encoded as genotypes, and phenotypes are decoded by attractor selection using noise-induced fluctuations. Through evaluation, we show that the attractor selection by individuals leads to the discovery of appropriate solutions faster than when using neural networks.