Latency Reduction in CloudVR: Cloud Prediction, Edge Correction

TL;DR

This paper introduces a cloud-edge collaborative method for VR rendering that predicts head movement on the cloud and re-renders on the edge, significantly increasing user capacity and reducing latency.

Contribution

It presents a novel hybrid cloud-edge approach for VR rendering that improves user capacity and prediction accuracy through specific prediction and normalization techniques.

Findings

Each edge server can serve up to 23 users concurrently.

Prediction accuracy is improved by using Mean Absolute Error loss and predicting acceleration.

Normalizing data by mean and standard deviation does not improve accuracy.

Abstract

Current virtual reality (VR) headsets encounter a trade-off between high processing power and affordability. Consequently, offloading 3D rendering to remote servers helps reduce costs, battery usage, and headset weight. Maintaining network latency below 20ms is crucial to achieving this goal. Predicting future movement and prerendering are beneficial in meeting this tight latency bound. This paper proposes a method that utilizes the low-latency property of edge servers and the high resources available in cloud servers simultaneously to achieve cost-efficient, high-quality VR. In this method, head movement is predicted on the cloud server, and frames are rendered there and transmitted to the edge server. If the prediction error surpasses a threshold, the frame is re-rendered on the edge server. Results demonstrate that using this method, each edge server can efficiently serve up to 23 users concurrently, compared to a maximum of 5 users when rendering the frame entirely on the edge server. Furthermore, this paper shows that employing the Mean Absolute Error loss function and predicting acceleration rather than velocity significantly enhances prediction accuracy. Additionally, it is shown that normalizing individual data using its mean and standard deviation does not yield improvements in prediction accuracy. These findings provide insights into optimizing VR headset performance through cloud-edge collaboration.