Analysis of DASH performance over time-varying end-to-end links

Abstract Video streaming today shows a widespread success, representing already the greatest portion of the Internet's IP traffic. The most common approach in today's networks is to adopt HTTP-based solutions, such as the Dynamic Adaptive Streaming over HTTP (DASH) protocol. This choice allows using general purpose web clients (web browsers) to access video streaming, while implementing a client-side dynamic rate adaptation of the streaming service. In fact, such HTTP-based approach must cope with intrinsic dynamism of the client's access network, which is a function of concurrent traffic from either the same user or other users sharing the same network resources. In addition, the network virtualization era fosters dynamic management of virtual resources that can lead to even further network variability, with significant changes of the end-to-end link characteristics in terms of bottleneck bandwidth and physical delay. Last but not least, the fruitful utilization of the satellite component in the future 5G networks implies to experience physical delay changes over a larger value range. Therefore, DASH flows will run on communication scenarios with higher and higher variability: this means that the client-side dynamic adaptation feature of DASH becomes a critical component of the whole service. This paper aims to investigate performance of two main DASH Adaptive Bit Rate (ABR) algorithm categories, throughput-based and buffer-based, over a simulated time-varying end-to-end link modeling the joint effects of continuous traffic and link changes. The goal is to provide statistical overview on the DASH ABR adaptation on challenging scenarios in order to draw some baseline recommendations for possible enhancements or modifications. The overall performance analysis is based on a customized script running in the Network Simulator 3 (NS-3).