Avaliação de escalabilidade e desempenho do plano de controle em redes definidas por software com OVN

Abstract

The adoption of Software-Defined Networking (SDN) in large-scale cloud environments presents significant challenges to the control plane. While the data plane has received substantial optimization through technologies such as DPDK, control plane performance—specifically the convergence latency between logical intent and physical realization—has emerged as a new critical bottleneck. Within the OpenStack ecosystem, Open Virtual Network (OVN) serves as the central orchestration component, translating Northbound logical configurations into Southbound OpenFlow flows. In legacy versions, the ovn-northd daemon performed a full recomputation of the network graph for every state change, leading to excessive CPU consumption and high latency in high-density port scenarios. This work utilizes the synthetic injection of network resources (Logical Switches and Ports) to quantify convergence time and computational resource usage. The experiments compare performance across different versions, evaluating the specific impact of the Incremental Processing mechanism. Results demonstrate that algorithmic optimization in the control plane reduces network provisioning time by orders of magnitude, thereby enabling scalability for thousands of virtual instances.