A Genetic Algorithm for Placing VNF Chains with Multiple Flavours

The Network Function Virtualisation (NFV) paradigm has revolutionised the way networks are designed, deployed, operated and managed by leveraging the flexibility, scalability and cost efficiencies of the cloud. The emergence of the Cloud-Edge Continuum allows network services, or parts of them, to be delivered close to the end user, especially when latency and throughput requirements are stringent. However, this perspective also poses several resource orchestration challenges, such as the need to cope with rapid changes in the infrastructure status, network and computing resource shortage, and diverse service requirements. In a previous paper, we demonstrated through simulations that a more flexible matching of network service requests with available resources can be enabled by the concept of multi-flavoured network services, i.e. services whose specifications include a full-fledged version and possibly alternative, less demanding versions (with less stringent resource requirements and/or fewer offered features). Since the VNF chain placement problem is known to be NP-hard, we propose a genetic algorithm-based metaheuristic to efficiently solve this variant of the VNF chain placement problem. Simulation results suggest that our genetic algorithm can achieve a profit improvement over a greedy solution of up to 8% in a 28-node topology and up to 6.6% in a 50-node topology.