From Audited Queue Regulation to Strategy-Level Reproducibility: Comparative Experimental Evidence from the Tb3-vntu and Tb5-tcdu Virtual 5G/O-RAN Testbeds

This paper connects two consecutively executed, evidence-audited experimental programmes in the Sci-O-RAN virtual 5G/O-RAN environment. An earlier Tb3-vntu study established bounded, causal regulation of a measured radio link control (RLC) queue through confirmed physical resource block (PRB) actuation; its 14-attempt PI replication population included 12 complete observations, 11 of which met seven predeclared pilot gates. The subsequent Tb5-TCDU programme preserved the native queue-feedback concept but addressed a different question: comparison of closed-loop Queue_PI with three static PRB allocations (13, 26 and 39 PRB), each scheduled for three repetitions. All 12 attempts are retained: 10 are complete and independently audited, one has 499 rather than 500 queue samples, and one terminated on a PRB-issue deadline. In two completed Tb5-tcdu PI sessions, mean confirmed allocation was 16.88 PRB, versus 26 and 39 PRB for fixed baselines, while all three strategies delivered the prescribed aggregate UDP payload throughput of 9.549824 Mbit/s without recorded packet loss. The corresponding mean raw queue during load was approximately 239 kB for PI but only approximately 1.2 kB for fixed 26 or 39 PRB. All three completed 13-PRB runs instead exhibited a repeated raw queue maximum of 4,313,895 B and 10.61–13.18% packet loss. Using an explicitly reconstructed trailing-five-sample metric, the two completed Tb5-tcdu PI runs had tracking RMSE values of 61.07 and 59.93 kB; the previous Tb3-vntu report gives 48.29 kB as the mean across its 12 complete PI replications. These values describe different experimental populations and must not be interpreted as a controlled cross-testbed effect. The contribution is an auditable strategy comparison, an explicit inter-testbed comparability contract, and evidence of a resource–queue trade-off, rather than a new PI law, a stability proof, or validated M2/M3 plant models.