Title: Concord: Airtime-Aware Contention Control for Taming Tail Latency from Wi-Fi Frame Bursting
Authors: Fengqian Guo, Siqi Wei, Sihao Miao (University of Science and Technology of China); Xinle Du (Tsinghua University); Hancheng Lu (University of Science and Technology of China)
Introduction
This paper studies how to reduce tail latency for interactive applications such as cloud gaming and extended reality (XR) in congested Wi-Fi networks. Although modern Wi-Fi offers PHY rates of up to 9.6 Gb/s and beyond, users in densely deployed WLANs can still experience unpredictable latency spikes of hundreds of milliseconds, even when their traffic rate is low. Existing approaches fall short for fundamental reasons: sender-side congestion control has little influence over channel-access delays at the IEEE 802.11 MAC layer; AP-side QoS mechanisms cannot reliably infer application intent from encrypted traffic such as QUIC; and EDCA provides only coarse-grained access categories, making it difficult to reconcile long bursts optimized for throughput with short bursts optimized for latency.
Key Idea and Contribution
The authors build Concord, an airtime-aware Wi-Fi MAC mechanism that uses burst airtime as an explicit control signal. Its central insight is that reducing a flow’s sending rate does not necessarily reduce its latency in a congested WLAN. Tail latency is primarily determined by MAC-layer contention dynamics and how long other transmitters occupy the channel. Therefore, controlling airtime consumption, rather than only regulating transmission rate, is essential for taming Wi-Fi tail latency.
Concord has two main mechanisms. First, it limits the duration of each burst after a transmitter wins channel access, preventing vendor-specific frame bursting from freezing the backoff counters of other devices for tens of milliseconds. Second, it scales the contention window according to the estimated airtime of each transmission grant: short bursts receive smaller contention windows, while long bursts receive larger ones. This design requires no traffic classification, client modification, or protocol-stack changes, and introduces only O(1) overhead per burst. The paper also develops an analytical model of DCF contention under burst grants and shows why the contention window is the key control knob for the tail of access delay.
Evaluation
The authors evaluate Concord on a commercial 802.11ax router platform under microflow, cloud-gaming, heterogeneous traffic, and legacy-device coexistence scenarios. With four saturated downlink contenders, Concord reduces the 99.9th-percentile delay of a 100-byte microflow from 298 ms under the IEEE baseline and 461 ms with vendor bursting to 42 ms. For cloud-gaming traffic, the tail latency falls from 441 ms to 92 ms, while starvation is reduced by up to 10 times. In a ten-flow heterogeneous workload, short-burst flows achieve a 29 ms 99.9th-percentile latency, compared with 390–500 ms for all flows under the baseline. Median starvation within 100 ms windows decreases from 21% to 2.1%, and RTS success rate improves from 48.7% to 86.5%. This result is significant because it shows that airtime-aware contention control can directly reshape MAC-layer access behavior and substantially improve the responsiveness of latency-sensitive applications without requiring changes to clients or encrypted traffic.
None.
Personal Thoughts
This paper identifies an important problem that is often obscured by the rapid growth of Wi-Fi PHY rates. I particularly appreciate its diagnosis that the main cause of tail latency is not simply excessive sending rate, but MAC-layer burst occupancy that freezes the backoff process of competing devices. The resulting design is simple and elegant: transmitters that consume more airtime should wait longer before accessing the channel again.
The evaluation is convincing because it covers both controlled microflows and realistic workloads such as cloud gaming, as well as coexistence with legacy devices. The AP-only deployment model also makes Concord practically attractive. An interesting direction for future work would be to study how its airtime-based policy interacts with multi-AP coordination, uplink contention, and emerging Wi-Fi scheduling mechanisms.