PPO-SL-QoS: Evaluation Dataset and Pre-trained Model for QoS-Priority SDN Scheduling

This dataset accompanies the paper "Architecture-Guaranteed QoS-Priority Bandwidth Protection in Software-Defined Networks via Proximal Policy Optimization and Sainte-Laguë Proportional Allocation." PPO-SL-QoS is a two-timescale SDN scheduling framework that combines Proximal Policy Optimization (PPO) with Weighted Sainte-Laguë (WSL) proportional allocation to guarantee QoS-priority bandwidth ordering under dynamic weight-change events. The PPO agent (~100 ms timescale) learns per-class bandwidth multipliers with a structural constraint α_gold > α_silver > α_bronze, while the WSL scheduler (~10 ms timescale) allocates slots proportionally using non-resetting counters that preserve convergence across weight-change events. This archive contains: 600 independent evaluation runs (6 scheduling methods × 5 traffic scenarios × 20 runs), including per-run JSON summaries and per-step time-series CSV filesAggregated statistics and statistical test results (Mann-Whitney U, Cliff's delta)Publication figures (bw_ratio heatmap, QoS priority proof, convergence profile, statistical dominance, ablation)Pre-trained PPO checkpoint (episode 20) and training curve Key result: PPO-SL-QoS achieves bw_ratio_gold of 1.393±0.043 (SC-2) to 1.403±0.049 (SC-3) with 95–100% priority ordering rate per scenario (SC-2: 100%, SC-3: 95%, SC-4: 100%, SC-5: 100%), versus 0% for all five baselines. Cliff's δ = 1.000 (p < 10⁻²⁷) against all baselines across pooled weight-change scenarios (N = 80 per method). Testbed: Mininet 2.3, Open vSwitch 3.1, Ryu OpenFlow 1.3, Fat-Tree k=4, 16 clients, 9 servers in 3 QoS groups. Code repository: https://github.com/anandahadi-usk/ppo-sl-qos-sdn