Experimental Study of Zero-Copy Performance for Immersive Streaming Service in Linux

With the development of network technology, many new services are emerging such as virtual reality, Augmented Reality and Holographic-Type Communication. In particular, immersive streaming service demand on the network to satisfy transmission speed, number of device connections and low latency. Even if such an immersive service need ultra-low latency, most of the research focuses on the reducing delay to the network. In other words, in order to satisfy end-to-end ultra-low latency, packet processing in the internal network layer of end devices is important. This paper compares and analyzes the high-speed packet processing technologies such as DPDK and XDP of the network layer in Linux. Our implementation and experiments shows the applicability of each technology with immersive service as an application.

I. Introduction

With advances in advanced communication technology, variety of immersive streaming services such as AR(Augmented Reality), VR(Virtual Reality) and HTC(Holographic-Type Communication) are emerging [1], [2]. However, such an immersive streaming services place significant demands on networking infrastructure in which many researchers are interested. These include the support for ultra-low delays, high bandwidth and dynamically adapt multiple data streams[3], [4]. Especially, In case of immersive streaming services, delay is the most important. Delays are classified into propagation, queuing, and transmission delays. The propagation delay is the rate at which packets travel along the network line. Queue delay refers to the time the packet waits in the queue until it can be processed. The transmission delay is the time a network node transmits a packet to the network [5]. We describes techniques to reduce the transmission delay for truly immersive streaming service and measures the performance of each technique. View point of streaming service a method for transmitting data streams such as TCP(Transmission Control Protocol) and UDP(User Datagram Protocol) is required. Streaming that prioritizes quality over latency tend to select TCP, whereas the services that optimize real-time manner choose UDP. TCP based transmission prioritizes quality against latency, which incurs additional delay because continuous packet acknowledgment is required. Due to extra delay for acknowledgment, TCP is not suitable for real-time or interactive services. Conversely, UDP is more suitable for low-latency services without guaranteed quality [6]. For this reasons, the existing TCP and UDP as the transport protocol is not suitable for new streaming service. In other words, TCP is not able to guarantee performance and UDP has weak reliability.