Follow us on: GitHub YouTube Vimeo Twitter

Modeling and Optimization of Network Assisted Video Streaming

Publication date: 2020-10-21

Computer networks are shared by multiple users and devices. Traffic from multiple applications is combined on a single network link, dividing the bandwidth between these applications. Video streaming, one of the most popular applications on the Internet, requires a relatively large and stable share of the bandwidth to be able to deliver high quality video. The architecture and protocols used by the dominant streaming technology, HTTP Adaptive Streaming (HAS), have enabled the massive scale of video content delivery as we know it today. Yet they are not suitable for meeting the demanding bandwidth requirements in shared networks.

Failure to deliver high quality video streams reflects badly on content providers and network operators, who risk the loss of revenue due to unsatisfied users. Video delivery problems distract the user, cause annoyance, and may lead to users abandoning their streams. Many scientific efforts have been performed to improve the streaming experience by optimizing the internals of HAS players. Such optimizations focus on the adaptation algorithms in HAS players to overcome bandwidth estimation errors, but they overlook the crucial role of the network in providing an environment where HAS players can operate without difficulty.

In his PhD thesis “Modeling and Optimization of Network Assisted Video Streaming", CWI PhD student Jan Willem Kleinrouweler validates the hypothesis that the addition of a network-based control element that guides HTTP adaptive streaming clients improves the video quality and fairness between clients in shared networks. To this end, mechanisms and policies for bandwidth sharing between HAS clients were developed, analyzed and optimized. By combining empirical and theoretical research methodologies the thesis contains a broad understanding of network optimizations for video streaming. Real implementations and experiments in different networks were used to evaluate and improve the mechanisms for interaction between HAS clients and the network. Markov modeling was used to further understand the impact of sharing policies on the overall streaming performance.

During his PhD research, Kleinrouweler created a network-based control element, which supports HAS players to stream at the optimal bitrate given the current network state. An extensive series of experiments shows that this solution significantly improves the streaming performance by preventing video freezes, minimizing the number of quality switches, and providing a fair video quality to all players. In experiments with up to 600 concurrent HAS clients, the scalability and effectiveness of the control element are demonstrated. The thesis continues with a method for understanding and predicting the overall streaming performance given a bandwidth sharing policy in the control element using Markov models. In validation experiments, which compare model-based results with results from the experiments with the control element, the high accuracy of our model is shown.

Overall, it can be concluded that the approach of integrating video players and network elements improves streaming performance over existing approaches in an efficient way. Modeling video streaming of multiple players in varying network settings proves to be an effective method to get results faster and gain deeper insights.

More Information:

  • PhD Defense Jan Willem Kleinrouweler: Wednesday 28 October 2020, 15:45-17:00 at VU University Amsterdam.
  • Promotors: prof. dr. D.C.A. Bulterman (CWI/VU) and prof. dr. R.D. van der Mei (CWI/VU)
  • Co-promotor: prof. dr. P.S. Cesar (CWI/TU Delft)

The defense can be followed online via the YouTube channel of the VU Beadle’s office:

Relevant Publications:

Original article