5.2. Evaluation
The airtime fairness algorithm proposed in this paper is a QoS scheduling algorithm designed specifically for Wi-Fi. Unlike general Wi-Fi queue scheduling algorithms, QoS scheduling and queue scheduling serve distinct purposes but work together to enhance the performance and user experience of wireless networks. The primary goal of a Wi-Fi QoS scheduling algorithm is to manage and optimize the transmission of various types of traffic within the wireless network to meet the specific quality of service requirements of different applications.
A common Wi-Fi QoS mechanism is Enhanced Distributed Channel Access (EDCA), defined in the IEEE 802.11 standard. EDCA classifies traffic into four types: voice, video, background, and best effort. Each has different priorities and parameters to ensure better service for high-priority traffic. In contrast, Wi-Fi queue scheduling algorithms manage packet queues on access points or terminal devices, handling enqueue and dequeue operations, transmission resource allocation, and traffic control. Examples of common queue scheduling algorithms include Drop-tail [
24], RED [
25], and CoDel [
26], which determine the order and timing of packet transmission. In this study, the classic Drop-tail queue was utilized.
Download Scenario. We begin by comparing the performance of three QoS algorithms in a download scenario: the default QoS algorithm, general airtime fairness, and the airtime fairness algorithm proposed in this paper. Throughput results are tested under varying TCP ACK delay conditions, as shown in
Table 2.
The experimental results reveal that the default QoS algorithm in this scenario aligns with throughput fairness, as the throughput for the three stations is nearly identical. In contrast, general airtime fairness tends to favor stations with higher wireless rates, providing them with more transmission opportunities. This approach results in higher effective throughput for these stations, thereby increasing the overall throughput of the wireless link.
Our proposed airtime fairness algorithm, however, produces more balanced throughput among the stations by adjusting airtime calculations differently for each station. As a result, while the total throughput achieved by our algorithm exceeds that of the default QoS algorithm, it falls short of the throughput provided by general airtime fairness. We would like to further explain that better airtime fairness does not equate to higher total throughput. For instance, if all wireless channel resources are allocated to the fastest STA, the highest total throughput would be achieved. However, this is clearly not a reasonable approach.
Consistent results across experiments with different TCP delayed ACK parameters demonstrate that our QoS algorithm effectively adjusts the stations’ access opportunities to the wireless channel based on their airtime, thereby achieving airtime fairness. The specific airtime for each STA in this experiment is detailed in
Table 3.
The experimental results indicate that the airtime fairness index of the default QoS algorithm is quite high, even significantly surpassing that of the general airtime fairness algorithm. Moreover, the default QoS algorithm can be viewed as throughput fairness in this test scenario. This discrepancy arises because the general airtime fairness algorithm miscalculates the airtime for each STA, leading to a substantial deviation from the actual results.
In contrast, our proposed airtime fairness algorithm accurately calculates airtime, enabling the AP to achieve excellent airtime fairness through a QoS-based scheduling approach. The experimental results closely align with the theoretical optimal solution, which has a fairness index of 1.
TCP Congestion Control. Since the primary goal of this paper is to improve the TCP airtime fairness, it is essential to evaluate the performance of our QoS-based scheduling method under different congestion control algorithms. This is because congestion control is one of the core components of TCP. We conduct experiments in this download scenario, where we sequentially set the same congestion control algorithms (Reno [
23], Vegas [
27], Cubic [
28], and BBR [
29]) to the three STAs. All other experimental conditions are kept consistent with those previously described. The results are presented in
Table 4.
The experimental results show that our method is effective and demonstrates consistently high fairness when facing different TCP parameters and popular congestion control algorithms. The reason is that we have incorporated the TCP parameters that affect our QoS-based scheduling algorithm into the modeling analysis.
Upload Scenario. Next, we examined the impact of each algorithm on upload TCP flows using the network topology shown in
Figure 9, with the throughput for each STA detailed in
Table 5.
Building on our airtime analysis of TCP ACKs in
Section 4.3, we extended the AP-based airtime fairness to upload traffic. The comparison between the default QoS algorithm and the two airtime fairness approaches demonstrates that our algorithm effectively controls changes in upload traffic throughput at the AP, thereby achieving airtime fairness.
However, in this scenario, the default QoS algorithm only achieves throughput fairness when the delayed ACK is set to 1. In contrast, our algorithm consistently achieves airtime fairness across various network environments, leading to an improvement in the overall throughput of the wireless network. The airtime for each STA in this scenario is presented in
Table 6.
In the upload scenario, the airtime results for each station generally align with those observed in the download scenario. However, it is notable that when the delayed ACK is set to 2, the general airtime fairness algorithm produces significantly higher results compared to other scenarios. In contrast, our airtime fairness algorithm consistently achieves high fairness across different scenarios, owing to its precise calculation of each station’s airtime.
In summary, the results of all the experiments reveal that the general airtime fairness algorithm calculates transmission airtime based on frame size and wireless rate, which is then used in the scheduling algorithm. This approach allows stations with higher wireless rates to have a greater opportunity to occupy the wireless channel, leading to a positive correlation between the throughput of each node and its wireless rate, thereby increasing total throughput. However, this approach presents two main issues:
Inaccurate Airtime Calculation: As the modulation and coding scheme increases under the same wireless protocol, the wireless transmission rate speeds up, reducing the proportion of time needed for data transmission. When the wireless rate is high, the general airtime fairness algorithm’s airtime calculation only reflects a fraction of the total time cost. This method significantly underestimates the actual airtime for nodes with higher rates, erroneously allocating them more transmission opportunities, which results in a poor airtime fairness index.
TCP ACK Considerations: TCP data packets require the return of TCP ACKs. Given that the wireless channel operates in half-duplex mode, the airtime cost associated with ACKs should be included in the total airtime calculation for the corresponding TCP data frame. The airtime for a single TCP data frame should therefore include both the forward data packet and the reverse ACK. This total airtime should be pre-calculated during AP scheduling to inform the scheduling logic. Our algorithm better reflects the concept of airtime fairness by incorporating these considerations.