Computer Laboratory

Technical reports

An architecture for scalable and deterministic video servers

Feng Shi

November 1997, 148 pages

This technical report is based on a dissertation submitted June 1997 by the author for the degree of Doctor of Philosophy to the University of Cambridge, Wolfson College.


A video server is a storage system that can provide a repository for continuous media (CM) data and sustain CM stream delivery (playback or recording) through networks. The voluminous nature of CM data demands a video server to be scalable in order to serve a large number of concurrent client requests. In addition, deterministic services can be provided by a video server for playback because the characteristics of variable bit rate (VBR) video can be analysed in advance and used in run-time admission control (AC) and data retrieval.

Recent research has made gigabit switches a reality, and the cost/performance ratio of microprocessors and standard PCs is dropping steadily. It would be more cost effective and flexible to use off-the-shelf components inside a video server with a scalable switched network as the primary interconnect than to make a special purpose or massively parallel multiprocessor based video server. This work advocates and assumes such a scalable video server structure in which data is striped to multiple peripherals attached directly to a switched network.

However, most contemporary distributed file systems do not support data distribution across multiple networked nodes, let alone providing quality of service (QoS) to CM applications at the same time. It is the observation of this dissertation that the software system framework for network striped video servers is as important as the scalable hardware architecture itself. This leads to the development of a new system architecture, which is scalable, flexible and QoS aware, for scalable and deterministic video servers. The resulting srchitecture is called Cadmus from sCAlable and Deterministic MUlitmedia Servers.

Cadmus also provides integrated solutions to AC and actual QoS enforcement in storage nodes. This is achieved by considering resources such as CPU buffer, disk, and network, simultaneously but not independently and by including both real-time (RT) and non-real-time (NRT) activities, In addition, the potential to smooth the variability of VBR videos using read-ahead under client buffer constraints is identified. A new smoothing algorithm is presented, analysed, and incorporated into the Cadmus architecture.

A prototype implementation of Cadmus has been constructed based on distributed object computing and hardware modules directly connected to an Asynchronous Transfer Mode (ATM) network. Experiments were performed to evaluate the implementation and demonstrate the utility and feasibility of the architecture and its AC criteria.

Full text

PDF (11.3 MB)

BibTeX record

  author =	 {Shi, Feng},
  title = 	 {{An architecture for scalable and deterministic video
  year = 	 1997,
  month = 	 nov,
  url = 	 {},
  institution =  {University of Cambridge, Computer Laboratory},
  number = 	 {UCAM-CL-TR-438}