Practical Latency-Aware Scheduling for Low-Latency Elephant VR Flows in Wi-Fi Networks.

Virtual reality (VR) applications are increasingly popular. With high-quality video streams and interactive content, they require both low-latency and high-bandwidth performance demands on the communication from the edge-based VR server to the VR headsets. Although most VR headsets are equipped with dedicated wired or wireless modules connected to the VR server, using common Wi-Fi networks to support them can be a promising trend due to convenience and low cost. However, current Wi-Fi Access Points (APs) cannot meet latency demands of low-latency elephant VR flows, especially in traffic congestion cases. We thus design a practical Wi-Fi scheduling solution, designated as LAST-PQ (Latency-Aware Scheduler with Two-level Priority Queueing), to support VR flows at the Wi-Fi AP. It monitors the runtime latency performance of VR flows while prioritizing scheduling for urgent flows, whose latency demands are at risk of violation. We implement LAST-PQ in Linux on a commodity Wi-Fi platform using an open-source Wi-Fi driver; it is compliant to the current Wi-Fi scheduling framework. The evaluation result shows that it can reduce latency by up to 79.89% in various congested scenarios; moreover, it consistently meets the latency demands of VR flows in cases of mobility at runtime.