Reliable and Real-time Content Streaming for Cloud and Edge Computing

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering


PhD Thesis Defence


Title: "Reliable and Real-time Content Streaming for Cloud and Edge Computing"

By

Mr. Ahmad ALHILAL


Abstract

The modern urban landscape is becoming increasingly connected, from distributed 
vehicular safety systems to in-vehicle centralized entertainment. These 
technologies face network challenges related to the high mobility of vehicles 
and people. Such mobility raises new challenges. On-board detection of unsafe 
driving activities lacks a holistic view of the road situation while 
cloud-offloaded detection faces scalability and context-awareness issues. In 
terms of in-vehicle entertainment, the connection to remote gaming and VR 
servers encounters variable bandwidth, latency, and packet losses that affect 
the gaming and VR experience.

This thesis presents four research contributions to provide reliable real-time 
content streaming for driving awareness(in the physical world), mobile cloud 
gaming (in the virtual world), and a reality-check of Metaverse between 
physical and virtual worlds.

The first contribution is CAD3, a distributed collaborative architecture for 
road-aware and driver-aware anomaly driving detection and real-time warning 
dissemination. CAD3 exploits the pervasive deployment of roadside units, and 
combines collaborative and context-aware computation with low-latency 
communication to detect unsafe driving behaviors and warn the drivers of nearby 
vehicles in real-time.

The second contribution is Nebula, an end-to-end cloud gaming architecture to 
minimize the impact of network conditions on the user experience. Nebula relies 
on a heuristic algorithm and end-to-end distortion model to dynamically adapt 
the video bitrate and redundancy based on the measured network conditions.

The third contribution is MERA, an edge-assisted learning-based end-to-end 
cloud gaming architecture to adapt the video bitrate to the network 
constraints. MERA relies on the transition, state-to-action mapping, then 
rewarding with a multi-objective reward function to maximize the user QoE.

With these contributions, end-to-end content streaming architectures are 
developed for time-critical applications to obtain minimal local latency, the 
time span between driving anomalous events and warning dissemination (CAD3) 
while obtaining high detection accuracy and system scalability, and between the 
user input and playing back the corresponding frame (Nebula and  MERA) while 
maintaining satisfactory video quality.

The fourth contribution is Metaversity, a reality check of a social VR 
opensource towards university-scale metaverse. This study dissects Mozilla Hubs 
and investigates the underlying networking and system scalability.


Date:			Tuesday, 26 July 2022

Time:			4:00pm - 6:00pm

Zoom Meeting:
https://hkust.zoom.us/j/97548494014?pwd=TGdxVEx1dlJqdDh5bTVMa1NZOEdPZz09

Chairperson:		Prof. Yang LU (ECON)

Committee Members:	Prof. Pan HUI (Supervisor, EMIA)
 			Prof. Tristan BRAUD (Supervisor)
 			Prof. Gary CHAN
 			Prof. James KWOK
 			Prof. Lilong CAI (MAE)
 			Prof. Georgios SMARAGDAKIS (TU Delft)


**** ALL are Welcome ****