Adaptive and High-Quality Multimedia Streaming

各种形式的多媒体流媒体（包括YouTube和Netflix）占北美互联网高峰流量的58.6%。除了不断增长的需求外，由于不同的连接类型（如WiFi、有线局域网和蜂窝）以及终端用户设备上不同的硬件规格，导致网络条件的异构和波动，对多媒体流媒体服务的适应性提出了挑战。虽然有解决方案，如动态自适应流媒体over HTTP （DASH）和苹果的HTTP Live Streaming (HLS)，以满足日益增长的需求，使多媒体流媒体服务更具适应性，但它们都没有充分利用所有可用的网络资源，以尽可能高的质量传输视频。例如，即使从服务提供商到终端用户设备的可用带宽为400kbps，足以提供比标准清晰度质量（以192kbps比特率）更好的视频，也只能传输标准清晰度视频，因为在标准清晰度和高清晰度（以2mbps比特率）之间没有中间质量水平。提供一个连续的视频质量选择方案是具有挑战性的，因为转码（对于每个质量级别）是一个耗时和CPU密集型的过程。如何根据网络的动态情况选择合适的质量水平也是一个挑战。

Multimedia streaming in all forms (including YouTube and Netflix) constitutes 58.6% of peak Internet traffic in North America. In addition to the soaring demand, the heterogeneous and fluctuating network condition, due to different connection types (e.g., WiFi, wired LAN, and cellular), and diverse hardware specifications on end-user devices challenge multimedia streaming services to be more adaptive. Though there are solutions, such as Dynamic Adaptive Streaming over HTTP (DASH) and Apple's HTTP Live Streaming (HLS), to meet the increasing demand and to make multimedia streaming services more adaptive, none of them fully utilizes all available network resources to deliver the video in the best possible quality. For example, even if the available bandwidth from a service provider to an end-user device is 400 Kbps, sufficient to serve a video better than the standard definition quality (at 192 Kpbs bitrate), only standard definition video will be delivered since there is no intermediate quality level between standard definition and high definition (at 2 Mbps bitrate). It is challenging to provide a continuous video quality selection scheme since transcoding (for each quality level) is a time consuming and CPU intensive process. It is also challenging to select the appropriate quality level subject to the dynamic network condition. In this research program, I propose a new design for multimedia streaming that streams high-quality (approximating the best possible quality level) video. The new design introduces cloud-based video transcoding algorithms to prepare videos at different quality levels. It also includes a new model defining quality levels, based on which I will design quality adaptation algorithms offering fine-granular video quality selection. Finally, the research will develop new streaming algorithms and protocols incorporating the new transcoding algorithms and quality adaptation algorithms. In contrast to existing streaming solutions, the new design will provide a custom-fit streaming service that streams videos in the best possible quality subject to the network condition and status of end-user devices. All research results will be implemented and evaluated with standard video players in real clouds and networks. This research program will deliver an innovative content-aware streaming design offering adaptive and high-quality streaming services. It will lay the foundation for advancing the QoS (Quality-of-Service) of multimedia streaming services in Canada. It will provide solutions to cope with increasing demand for high-quality streaming service in major carrier networks across Canada (such as Shaw, Rogers, and Telus), and will inspire new streaming services for streaming content providers (such as TV stations and Netflix-like services).