SLO-aware dynamic self-adaptation of resources

Abstract

Cloud computing and Network Function Virtualization (NFV) are two complementary technologies. Virtual network functions (VNFs) provided by NFV are connected in the form of service function chains (SFCs) and typically hosted on the cloud. Dynamic resource adaptation in NFV-cloud settings remains a challenging research problem. VNF resources can be adapted by performing either vertical scaling (VS), horizontal scaling (HS), or Migration (M). Deciding on the optimum strategy among these three approaches (VS, HS, M) may face several challenges, including the dynamicity of the cloud environment; the sheer multiplicity of SFC topologies (e.g., linear, or non-linear SFCs); potentially conflicting optimization objectives, and the substrate network configuration. Considering the challenges introduced, we propose decision-making algorithms that make the best adaptation decisions for the SFCs dynamically, while balancing a set of cost functions, such as energy consumption, resource utilization, and Service Level Objective (SLO) violation. We first formulate the problem as an integer linear programming (ILP) model to compute the optimal solution. Then, because solving an ILP model is time-consuming, we adopt multi-objective metaheuristic algorithms based on Non-dominated Sorting Genetic Algorithm (NSGAII), Chemical Reaction Optimization (CRO), Binary Particle Swarm Optimization (NBPSO), and the combination CRO-NBPSO to solve this problem. Experimental results demonstrate the effectiveness of the proposed meta-heuristic algorithms in reducing the end-to-end latency while achieving performance similar to optimal solutions in terms of resource utilization and energy consumption.