<![CDATA[Selecting an industrial communication protocol is usually treated as a single, plant-wide decision, although factory networks host traffic classes whose timing requirements differ by more than three orders of magnitude. This paper presents a reproducible, workload-aware benchmark of three widely deployed protocols—Modbus TCP, OPC UA PubSub over Time-Sensitive Networking (TSN), and EtherCAT-across five representative workloads ranging from 250 µs synchronized servo control to 1 s condition monitoring. Stochastic latency models for each protocol were calibrated to published measurement studies and specifications and evaluated through a Monte Carlo experiment of 30 independent replications × 4,000 cycles per protocol-workload pair (1.8 million simulated cycles in total). Performance is quantified by on-time reliability, jitter, tail latency, and a composite score combining reliability, determinism, and deadline headroom. EtherCAT delivered 100% on-time reliability with sub-microsecond jitter on every workload. OPC UA PubSub achieved ≥99.87% reliability for deadlines of 1 ms or longer but only 75.3% at 250 µs. Modbus TCP was unusable below 2 ms yet reached ≥99.8% at 10 ms and above (average 34.5). The ranking is statistically significant (Kruskal–Wallis, p < 10⁻¹⁵ for every workload) and robust to changes in composite-score weights. Sensitivity analyses quantify the degradation of OPC UA PubSub with TSN hop count and of Modbus TCP with background load. The results support a segmented architecture-EtherCAT for motion control, OPC UA PubSub/TSN as the cell-to-enterprise backbone, and Modbus TCP for legacy, slow telemetry-and provide an open, recalibratable methodology for site-specific protocol selection.]]>
