EDSP-Edge: Efficient Dynamic Edge Service Entity Placement for Mobile Virtual Reality Systems

As one of the significant supporting technologies for mobile virtual reality (MVR), computer vision is latency-sensitive and always requires real-time response and accurate object analysis. However, the limited computational resources of mobile devices lead to high service delay and low analysis quality, resulting in poor quality of service (QoS). By placing the edge service entities (SEs) of the video tasks on the edge server close to the mobile users, a satisfactory QoS can be obtained for MVR systems. Most of the previous works are restricted to optimizing QoS through service placement, while ignoring the key impact of the network access point and video frame resolution selections on QoS. In this paper, we propose an edge service entity placement model, which aims to jointly optimize service delay and analysis quality for MVR systems. Specifically, we design an efficient dynamic edge service entity placement scheme (EDSP-Edge) based on the block coordinate descent theory, which dynamically determines the selection strategies of network access points, service entities and video frame resolutions for users to effectively improve the QoS. We theoretically analyze the performance of EDSP-Edge and get the gap between EDSP-Edge and the optimal performance. Finally, extensive real-data driven simulations are conducted to show that the EDSP-Edge performs close to the optimal scheme and achieves at least 22% performance improvement compared with previous works.

I. Introduction

With the explosive growth of mobile and Internet of Things (IoT) devices, applications running on mobile phones, tablets, virtual reality glasses and other terminal devices are becoming more and more popular. Some latency-sensitive applications such as virtual reality (VR), augmented reality (AR), and multiplayer online games require low delay to satisfy users’ demand for quality of service (QoS). However, terminal devices have limited storage space and computational resources and are always far from cloud servers. The high and frequently unpredictable delay between the cloud and the end user makes the traditional cloud computing paradigm inapplicable to mobile virtual reality (MVR) systems [1], [2].