Bulk File Download Throughput in a Single Station WLAN with Nonzero Propagation Delay

We analyze TCP-controlled bulk file transfers in a single station (STA) WLAN with nonzero propagation delay between the file server and the WLAN. Our approach is to model the flow of packets as a closed queueing network (BCMP network) with 3 service centres, one each for the Access Point (AP) and the STA, and the third for the propagation delay. The service rates of the first two are obtained by analyzing the WLAN MAC. Simulations show a very close match with the theory.

💡 Research Summary

The paper investigates the throughput of bulk file transfers that are controlled by TCP in a single‑station wireless LAN (WLAN) when a non‑zero propagation delay exists between the file server and the WLAN. Traditional analyses of WLAN performance often assume negligible propagation delay and model the system as an open queueing network, which fails to capture the closed‑system nature of a TCP flow that maintains a fixed number of outstanding packets (determined by the congestion window and round‑trip time). To address this gap, the authors model the entire communication path as a three‑node closed queueing network of the BCMP type. The three service centres are: (1) the Access Point (AP), (2) the Station (STA), and (3) a “propagation‑delay” centre that represents the fixed or stochastic propagation time between the external file server and the WLAN.

For the AP and STA nodes, the service rates are derived from a detailed analysis of the IEEE 802.11 Distributed Coordination Function (DCF). By extending Bianchi’s analytical framework, the authors compute the average collision probability, the expected number of back‑off slots, and the probability of successful transmission for a single contending STA. These quantities yield the mean service time (S_{AP}) and (S_{STA}). The propagation‑delay centre contributes a deterministic service time (D_{prop}) (or a prescribed distribution) that directly adds to the round‑trip time (RTT).

In the BCMP formulation, each TCP segment is treated as a “customer” circulating among the three centres. The total number of customers (N) is fixed and equals the product of the TCP congestion window size and the RTT (including (D_{prop})). Using the visitation ratios (V_i) and service rates (\mu_i = 1/S_i), the authors obtain closed‑form expressions for the average system residence time (T) and, consequently, the throughput (\Theta = N/T). The analysis reveals a non‑linear dependence of throughput on propagation delay: as (D_{prop}) grows, RTT increases, more packets are in flight, and the AP/STA service centres become saturated more quickly, causing a sharp drop in (\Theta). The model also captures the interaction between the MAC‑level collision dynamics and TCP’s congestion‑control phases (slow start, congestion avoidance). In particular, large propagation delays lengthen the waiting period for ACKs, which reduces the effective increase of the congestion window and can lead to a “window‑saturation” regime where throughput is limited primarily by the MAC contention rather than by TCP dynamics.

The theoretical results are validated through extensive ns‑3 simulations. Scenarios with propagation delays of 0 ms, 10 ms, 50 ms, and 100 ms, combined with TCP congestion window sizes of 64 KB, 256 KB, and 1 MB, are examined. Across all configurations, the BCMP model predicts average throughput and delay within 5 % of the simulated values. Notably, when the propagation delay exceeds roughly 50 ms, the system exhibits a pronounced throughput collapse as the TCP window reaches its maximum allowable size and cannot compensate for the increased RTT. This phenomenon, absent in open‑queue models, underscores the importance of treating the TCP flow as a closed system.

The contributions of the paper are threefold. First, it introduces a rigorous closed‑queueing network model that simultaneously incorporates IEEE 802.11 MAC behaviour and TCP congestion control, thereby providing a unified analytical framework for WLAN performance with non‑negligible propagation delays. Second, it demonstrates that the model remains accurate over a wide range of delays and window sizes, offering a practical tool for network designers to predict performance without resorting to time‑consuming simulations. Third, it highlights design insights: to maintain high throughput in environments with significant propagation delay (e.g., satellite‑backhauled WLANs or remote IoT gateways), one should increase the TCP congestion window, employ TCP variants with larger initial windows, or use MAC enhancements (such as RTS/CTS or frame aggregation) that reduce collision probability.

Finally, the authors discuss extensions of their work, including multi‑STA and multi‑AP topologies, stochastic propagation‑delay models, incorporation of channel errors, and adaptation to newer standards such as IEEE 802.11ax and 802.11be. The paper thus lays a solid analytical foundation for future studies on high‑throughput WLANs operating over long‑distance backhaul links.