Why TCP Reno Collapses over Noisy Wireless Links: A Simulation, Analytical-‎Model and Emulation-Based Evaluation

Purpose: To quantify how increasing channel noise, expressed as packet drop probability and the equivalent bit error rate (BER), degrades TCP Reno during a bulk FTP transfer, and to identify the mechanisms behind the degradation.

Materials and Methods: A client–IP cloud–server topology was simulated in OPNET Modeler 14.5, and a 10 MB file was downloaded under five drop probabilities from 0% to 10% (BER up to about 10⁻⁵). Results were compared with a published reference study, with closed-form expectations for datagram count, retransmissions and throughput bounds, with the Mathis and Padhye steady-state models, and with a virtualised Linux testbed.

Results: Response time stayed at 10.08 s up to 0.1% drop, rose 2.6-fold at 1% (26.01 s) and 191-fold at 10% (1,926.5 s), while effective goodput fell from about 992 kB/s to 5.2 kB/s. At 10%, 1,015 retransmissions were observed against 761 expected from independent losses. The Mathis model matched the 1% case within 18% but overestimated the 10% goodput nineteen-fold.

Conclusions: Degradation is strongly nonlinear and is dominated by window collapse and retransmission timeouts rather than by the raw number of lost packets. Default TCP Reno should not be used on links with drop probabilities of about 1% or more without link-layer recovery or loss-tolerant alternatives. (J Multidiscip Eng 2026;1:94-101)‎