Computer Science Department SAL 300
Phone: (213) 740-8162
Fax : (213) 740-5807
URL : http://infolab.usc.edu/
Senior Investigator: Prof. Roger Zimmermann
Ph.D. Students: Didi Yao, Farnoush Banaei-Kashani and Mehrdad Jahangiri
Project Award Information
Title: An Infrastructure for the Real-Time Delivery of High Quality Continuous Media from Distributed Scalable Servers to Multiple Clients
Continuous Media Servers, Real-time Streaming, Scalable Servers, Distributed Servers, Multimedia Infrastructure
Several large-scale IT applications are enabled by the design, implementation and evaluation of a scalable real-time streaming media architecture. Large volumes of real-time data are stored, maintained, and retrieved online. Popular examples of real-time media include video and audio, while less familiar examples are haptic and avatar data. The real-time end-to-end delivery of very high quality (megabits per second) media cannot be supported with the current Internet-based infrastructure due to the lack of a scalable server architecture as well as a missing global resource management protocol.
We are in the process of completing the design and implementation of our Yima streaming architecture, which consist of 1) a scalable, high performance, and real-time media server, 2) real-time network streaming paradigm and 3) several video and audio clients. In addition, we are working on the design and implementation of GMeN (Global Media Network), a server-level distributed streaming architecture for cost-efficient delivery of continuous media to geographically distributed clients. Also, with P2PS (Peer-to-Peer Streaming) we investigate the client-level distribution of service that pushes the streaming task to the edges/users.
The impact of this architecture is on both the server design as well as the online multimedia content. Furthermore, the realization of such an infrastructure is enabling large scale applications such as video-on-demand, news-on-demand, distance learning and scientific exploration and visualization. These applications, in turn, will be promoting teaching, training, and learning as well as enhancing scientific and technological understanding.
The following is a list of publications that are directly related to the topic of this grant and within which this ITR grant has been acknowledged. Several other research publications that have acknowledged this grant but are not directly related can be found at: http://infolab.usc.edu/Publication.html
The goal of this project is to design and develop an architecture for real-time storage and playback of several continuous media streams through heterogeneous, scalable and distributed servers utilizing IP networks (e.g., Internet). This architecture enables applications such as video-on-demand, employee training and distance learning on a large scale.
The main component of our architecture is a scalable media server, termed Yima. Yima is based on a cluster design comprised of multiple nodes of off-the-shelf personal computers resulting in both a scalable and cost-effective implementation. The network component of our architecture provides a selective retransmission scheme that improves the rendering quality of the transmitted streams with real-world networks that suffer from occasional packet loss. This scheme is shown to be more effective for real-time media as compared to conventional UDP and TCP schemes. Yima's SCADDAR algorithm allows incremental storage system growth while (a) the system continues operation, (b) the minimal amount of data is reorganized, and (c) the system continues to be load balanced. Finally, we are investigating a distributed architecture that consists of several Yima server nodes inter-connected through a controlled network. We have discovered that a decentralized management of the media objects (e.g., video clips), as opposed to a centralized (e.g., directory-service) design, results in more efficient and robust architecture. This fully decentralized design is also consistent with the recently popular peer-to-peer architectures. We have developed a working prototype of Yima that is designed to be media independent to stream media types such as MPEG-4, uncompressed immersive audio, panoramic (360 degree) video and High Definition TV. Our Multi-Threshold Flow Control (MTFC) algorithm allows the smoothing of variable bit-rate media transmissions without a priori knowledge of the actual bit-rate. It improves resource utilization and can be applied to (a) live streams and (b) stored streams without requiring any server side pre-processing. We have also shown the ability of Yima to allow multiple streams be rendered in tight synchronization over the public Internet. For example, panoramic video streams that are comprised of five individual streams can be displayed frame-locked.
Yima distinguishes itself from other similar efforts due to its: 1) independence from media type, 2) frame/sample accurate inter-stream synchronization, 3) compliance with industry standards (e.g., RTSP, RTP, MP4), 4) selective retransmission protocol, 5) scalable design, and 6) multi-threshold buffering to support variable-bitrate media (e.g., MPEG-2).
A video demonstration of Yima
is available on the project
web site. Several publications are also available as outlined in the
"Publications and Products" section.
A growing number of immersive and multimedia applications store, maintain, and retrieve large volumes of real-time data where the data is required to be available online. We denote these data types collectively as "continuous media", or CM for short. Continuous media is distinguished from traditional textual and record-based media in two ways. First, the retrieval and display of continuous media are subject to real-time constraints. If the real-time constraints are not satisfied, the display may suffer from disruptions and delays termed hiccups. Second, continuous media objects are large in size. A two-hour MPEG-2 video with a 4-megabit per second (Mb/s) bandwidth requirement is 3.6 gigabyte (GB) in size. Popular examples of CM are video and audio objects, while less familiar examples are haptic, avatar and application coordination data. We focus on those CM objects that are of very high quality requiring bandwidths in the order of megabits per second.
Some of the resource reservation and path selection protocols designed for our fully decentralized video servers are applicable to the peer-to-peer architectures (refer to the TPDS’02 paper). Our approach for scaling up the number of disk nodes, SCADDAR, can be adapted for the design of expandable storage systems for data streams (refer to the ICDE’02 paper).