Interaction of Traffic Load and Bottleneck Capacity on QoS Stability in Software-Defined Networking

Quality of Service (QoS) in software-defined networking (SDN) is usually evaluated by varying one factor at a time, so it remains unclear whether the effect of traffic load depends on bottleneck capacity. This study examines how offered load and bottleneck capacity interact to affect QoS stability in an emulated SDN. A dumbbell topology was built in Mininet with Open vSwitch and an os-ken controller installing static OpenFlow 1.3 rules. A full factorial design combined four offered loads (30, 60, 90, and 120% of capacity) with three capacities (5, 10, and 20 Mbps), each repeated in 10 independent sessions. Throughput, round-trip time (RTT), jitter, and packet loss were analysed with two-way ANOVA and the aligned rank transform, and queue counters confirmed that losses occurred at the bottleneck. Load dominated throughput and loss: delivery remained ≥ 98.6% up to 90% load and fell to about 80% at 120%. RTT showed a significant load–capacity interaction (p < 0.001): capacity had no significant effect up to 90% load, but at 120% mean RTT reached 93.3, 47.4, and 22.4 ms for 5, 10, and 20 Mbps. Between-session RTT variability peaked at 90% load, indicating that QoS instability emerges near saturation, before packet loss becomes persistent.