MIME-Version: 1.0 Server: CERN/3.0 Date: Wednesday, 20-Nov-96 18:57:08 GMT Content-Type: text/html Content-Length: 10260 Last-Modified: Friday, 31-May-96 14:15:26 GMT
Our research on storage systems is directed towards building the Zeno distributed video file server. The Zeno architecture uses a network of workstations connected by a generic local area network (e.g., an ethernet), a common environment in computing research laboratories. Each workstation can act simultaneously as both a client and a server of continuous media data. As a client, a workstation plays video stored on one or more servers. As a server, a workstation is a file server for video data. Each client can receive video stored at several servers, and each server can service several clients. Compared with large centralized servers, the advantages of this design are:
Our research on communication systems is centered around best effort delivery protocols. Such protocols are built on existing network protocols and, in contrast to many other research efforts, do not need to reserve network resources to establish a connection. Resource reservation protocols are well suited to the national communication infrastructure where users can be charged on a per call basis for bandwidth and connections, but are poorly suited to network environments where the network is a shared resource equally accessible by all researchers. Our approach is appropriate for the latter environments, commonly found in research laboratories. The communication protocol we have developed, called Cyclic-UDP, is built on top of the UDP datagram protocol, and is designed to transport audio and video data in playback applications in local, metropolitan, and wide area networks. Cyclic-UDP is used by the Zeno file server to deliver audio and video data to clients. A paper describing cyclic UDP is available online , as well as the slides from a research talk.
Our research on processing video data has been two-fold. First, we are developing algorithms to process video data in the compressed representation. Processing video without decompression leads to dramatic speed-ups in processing performance since it both removes the time-consuming processes of compression and decompression and reduces the amount of data that must be processed. Experiments with an implementation of these ideas on JPEG compressed image data indicates that the data can be processed one to two orders of magnitude faster than what was possible with previous approaches. We are currently extending these ideas by parallelizing the algorithms using networks of workstations and by developing a method for transcoding video in software. In video transcoding, video is translated from one compression format to another, a useful operation for video file servers that must service heterogeneous clients. A paper describing compressed domain processing is available online.
Our research on video processing is also exploring methods to simplify experimentation with video processing by developing a programming language where video is a first class data type. This language, called Rivl (pronounced "rival"), allows video processing effects to be specified independent of the resolution and format of the source material. The language does for video what Postscript did for text and graphics: it provides a resolution independent method for specifying video processing. Thus, the same program can process low quality QuickTime video very quickly while editing decisions are made, and then be used to format a high quality finished product off-line, in much the same way that Postscript can be previewed on a workstation at low quality, then sent to a 2600 dpi printer for camera ready copy. A paper describing RVL is available online.
A talk that reviews our research on video processing, both compressed domain processing and Rivl, is available online.
The common theme of all these efforts is to provide tools to make video
usable in our research environment.
A talk that reviews this research
is also available online.
At Cornell University, I teach an undergraduate course (CS 314) on computer
architecture, and a graduate course (CS
631) on multimedia systems.
Teaching
Selected Publications
Selected Research Talks
Misc Links