The Internet has enabled the television revolution. New uses have encouraged the adoption of streaming as well as the development of OTT (Over the Top) from which several video leaders are benefiting. This has contributed to the popularization of VOD (Video on Demand) services. Video content now accounts for a major share of Internet traffic. For example, the Netflix's video traffic would alone represent 71% of the bandwidth consumed in the United States at peak times, twice as much as in 2011. On average, Netflix consumes 35.2% of the bandwidth, followed by YouTube with 17.5% and Amazon Prime Video at 4.3%. Managing the explosion of video traffic is a challenge that arouses the interest of several economic and technological players. One can further observe the development of emerging technologies such as UltraHD, 4K and HDR (High Dynamic Range). The optimal use of the available bandwidth is therefore a crucial topic for the major players in video streaming.
Protocol solutions, content encoding and compression techniques have been proposed to improve the fluidity and quality of reading. A typical example is adaptive streaming, which dynamically adapts the quality of the video to the available bandwidth.
The content broadcaster must assume the significant bandwidth costs from the ever increasing video quality, especially during peak periods. The legacy distribution system for video content distribution over the Internet is based on a Client/Server (C/S) model where the servers are provided by a third- party actor, the (Content Distribution Network) CDN provider. In addition to their cost, one can notice the difficulty of CDNs to follow the demand in the event of poor audience estimations. An alternative solution is Peer-to-Peer broadcasting, where the content is exchanged among the machines of the users in the session, which reduces costs and improves streaming quality.
Peer-to-Peer (P2P) networks play a key role in today's world of telecommunications. This role has materialized with the advent of P2PWeb type protocols 1 which combine the Peer-to-Peer paradigm with the traditional C/S model. The integration of these new networks has been further strengthened with the arrival of WebRTC standards and the legal use of P2P through the broadcast of audiovisual content (Live and VoD) on the Internet by quite a few SVoD platforms like Easybrocast. The P2PWeb protocol is a real hybrid protocol insofar as it uses both the C/S (with the CDN) and P2P models simultaneously. The advantage of this type of protocol is to facilitate administration while guaranteeing scalability. In these types of networks, the information concerning the broadcast session (content viewed, viewing time, quality viewed, geolocation, etc.) is also decentralized. Some of the information is on the content provider's server. The other part is distributed over all the peers which participate in the distribution of this content. A third part is on the servers, called managers, that act as a directory to connect the peers to each other. While efficient, the management of these so- called hybrid networks is complex. The latter stems from the nature and form of dissemination. A research work carried out at Easybroadcast 2,3,4, (PhD thesis defended October 2021), has made it possible to improve the quality of viewing, optimize the selection of Viewers (Peers) and the bandwidth consumed for diffusion. The optimization carried out in this work was essentially based on the network parameters and geolocation of the Viewers.
The objectives of this thesis will be to initially propose a method for retrieving hardware information (CPU, RAM, network card, etc.) from the platforms used to view content. The next step will to rip the full benefit of this information to allow dynamic operation of the P2PWeb protocol depending on the power of the platform. This would imply limitations in the number of connections (to other viewers) to open and an optimization of the resources to be exploited on the user's platform to ensure a better quality of service with the most ecological consumption of the resources of its platform. The candidate must also analyze the different combinations concerning the broadcasting platforms, players (Dash/HLS) and networks (WiFi, 4G, etc.) and establish a summary concerning the users (who? watches what? ...) to propose an optimized management approach. The exploitation of the different information will be the master plan of the broadcaster to reduce the cost of its bandwidth as well as maximize its audience. This “feedback” system that we want to explore should be correctly processed to avoid any risk of intensifying the phenomena of grouping (clustering) not only by group of the same network power but also by group of the same type of distribution platform. Data science and machine learning techniques will play a key role in the exploitation of this data.
Funding category: Cifre
PHD title: Doctorat d'Informatique
PHD Country: FranceOffer Requirements Specific Requirements
- Python 3.5 language, Python frameworks (like PyCharm, Jupiter Notebook, Spyder, Conda)
- Deep Learning Libraries (like TensorFlow, Keras)
- Other IT skills such as: networks and system (Unix, typically), knowledge of OOP (like Java), Agile,
- Machine learning and data science (namely neural network theory)
- Computer network control plane (algorithms and protocols), namely in Video Streaming and
Content Distribution NetworksContact Information