Seminar by Dr. Abdelhak Bentaleb (National University of Singapore)

Title: Enabling Optimizations of Quality of Experience in Internet Video Services.  

Abstract:

The last year has seen a significant transformation of many activities as people reinvent their way of life given constraints such as social distancing and remote work. In the short span of seven months, Zoom conference platform usage grew 30x from 10 million daily meeting participants in December 2019 to 300 million in June 2020. In addition to the surge in video conferencing, other types of media streaming such as live and on­ demand streaming services and immersive media applications have also grown tremendously. A recent report from Convivia shows that – On a global scale, streaming jumped more than 20% as compared to two weeks prior, daytime viewing jumped nearly 40% as compared to two weeks prior, 5-20 % increase in video streaming services subscription, in March 2020. Live streaming services, such as Twitch and YouTube Live have graduated from early experiments to highly profitable busi­nesses serving millions of users, while immersive media applications have also started to show significant potential thanks to technological advancements, e.g., the introduction of affordable 360-degree cameras, virtual and augmented reality (VR and AR) displays and apps, and evolving use cases, e.g., such as new forms of social media interactions.

With the dramatic growth of Internet video streaming traffic, there has been a practical challenge for Internet service providers (ISPs) in meeting the demands and requirements of existing content providers while maintaining a high profit. These content providers are relying on user engagement to generate revenues, maintaining high end-user Quality of Experience (QoE) has become crucial to ensure high user engagement. For instance, one short rebuffering event leads to 39% less time spent watching videos and causes significant revenue losses for ad-based video sites. On the one hand, this is challenged by the users' request for high-quality video content without rebuffering. On the other hand, the best-effort network infrastructure is suffering from sudden and high fluctuating bandwidth due to uncertain characteristics of bandwidth and the congestion on the paths. Instead of the traditional video streaming solutions, the new worldwide video delivery system HTTP Adaptive Streaming (HAS) is rapidly being deployed and is becoming the de-facto solution for Over-The-Top (OTT) adaptive video streaming services over the Internet. 

Despite increasing expectations for high QoE, existing HAS-based solutions have limitations to achieve the QoE needed by today's video services. They either require costly re-thinking of the network core or use suboptimal endpoint-based protocols and schemes to react to the dynamic Internet performance based on limited knowledge of the network. In this talk, I will describe how we can overcome those limitations and optimize end-user QoE for video streaming services. 

• First, I will present a novel suite of adaptive bitrate (ABR) solutions for on-demand video services which are inspired by the recent success of emerging and state-of-the-art techniques in many fields of computing. These solutions can be classified into three main categories: centralized-, distributed-, and learning-based. The centralized solutions benefit from the network management and resource allocation capabilities of Software Defined Networking (SDN) to guide video players in meeting the requirements of end-users. The distributed solutions use queuing theory, game theory and the consensus mechanism to provide efficient cooperation and collaboration between video players in a distributed way and without introducing explicit communication overhead that impacts end-users' QoE negatively. The learning-based solutions leverage the power of online reinforcement learning to learn the best bitrate decisions and achieve high QoE. In this regard, I will present three solutions that I have designed during my Ph.D: SDNDASH (centralized), GTA (Distributed), and  ORL-SDN (Learning).

• Second, I will present the main challenges in low-latency live (LLL) streaming and then present one of my recent solutions that tackles these challenges and strives to enhance the LLL streaming performance. The proposed solution re-visit and extends several important components (collectively called Low-on-Latency-plus, LoL+) in adaptive streaming systems. LoL+ includes three essential modules: learning-based bitrate adaptation, playback control, and throughput measurement/prediction. LoL+ has been in­tegrated into the open-source reference player dash.js. The dash.js media player is used by many of the world’s largest stream­ing organizations who are members of the DASH Industry Forum (e.g., Netflix, Amazon, etc.). 

• Finally, I will conclude my talk with my future directions and research projects for the next five years in networked and media streaming systems fields, e.g., designing an end-to-end (from capturing to rendering) AI-powered framework for immersive media streaming over next-generation networks (5G and beyond).

Bio:

Abdelhak Bentaleb is a Postdoctoral Research Fellow in the Media Management Research Lab (MMRL) at the School of Computing, National University of Singapore (NUS), mentored by Prof Roger Zimmermann. His research interests are computer networking and video delivery optimization. His current research goal is to design and build practical machine/deep learning models for large-scale real-world networked systems. Specifically, it focuses on investigating a new paradigm based on practical machine/deep learning techniques for solving challenging QoE optimization and congestion control problems in video delivery systems. Rather than design fixed algorithms, the aim is to develop data-driven systems that can dynamically learn to optimize the performance on their own using modern machine and deep learning techniques (e.g., deep reinforcement learning). As a result of well understanding and optimizing the video streaming QoE as well as adjusting congestion control based on various requirements of video streaming applications and services.  He obtained his Ph.D. from NUS and his Ph.D. dissertation was about enabling optimizations of video delivery in HTTP Adaptive Streaming, received two prestigious awards: SIGMM Award for Outstanding Ph.D. Thesis and DASH Industry Forum Best Ph.D. Dissertation Award. Many papers resulting from his research have been published in top venues and he won several awards.