Memetic Algorithm for Multi-objective Virtual Network Slice Configuration

Through virtual network slicing, a physical network in 5G can support various vertical applications with different reliability requirements for the end-to-end latency. This paper models the problem of virtual network slice configuration (VNSC) as a multi-objective optimization problem (cost vs. reliability), considering a real-world scenario in which the latency distribution of network nodes is unknown. Instead of evaluating the network end-to-end latency by full Monte Carlo Simulation (MCS), we design a more efficient approximation evaluation method based on MCS and distribution fitting. Furthermore, based on NSGA-II, a multi-objective memetic algorithm for VNSC is proposed by introducing domain knowledge in both population initialization and chromosome mutation processes. The experimental studies on different network architectures demonstrate the superiority of the proposed method, i.e., the total amount of the MCS data can be reduced to 2%, and the fitness evaluation times can be reduced to 53.3%-86.4% on the slice instances of different scales without sacrificing slice performance.