Presentation at Dagstuhl Seminar 1999

Improving the Performance of Media Servers Providing Physical Data Independence—Problems, Concepts, and Challenges

Ulrich Marder

University of Kaiserslautern
P.O. Box 3049, 67653 Kaiserslautern, Germany
e-mail: marder@informatik.uni-kl.de

Online Presentation

The slides are available in HTML, PostScript (gzip), or PDF format.

This presentation is also provided as a SMIL-based slideshow with audio (RealPlayer® G2 required for playback).

Abstract

This presentation deals with media servers providing physical data independence. Today’s media servers—especially continuous media servers—usually do not provide physical data independence at all. One—if not the main—reason for this is performance. Physical data independence without optimization costs a lot of performance. Therefore, we are looking for a solution of this optimization problem.

The following is a typical scenario where physical data independence would be highly beneficial: There is global media data—often called media assets—stored in an MM-DBMS (which might be a part of a digital library). Lots of heterogeneous clients with different capabilities of storing, processing, and presenting media data are willing to access this MM-DBMS. Some only want to retrieve media objects for presentation or possibly printing. Others create or modify media objects. And again others create media objects by editing and combining existing ones. Assuming a  sort of unbalance between these applications sounds reasonable: Usually there will be many applications of the presentation type but few of the other types. Thus, one could think of optimizing the system for presentation. But unfortunately, the other applications often have much stronger quality and performance demands.

We show that there are considerable problems when attempting to provide physical data independence with a media server (or MM-DBMS). Such systems tend to require frequent format conversions inevitably resulting in bad performance. They may inadvertently lose data due to irreversible updates. And hiding the internal data representation from the client obviously means that all the strongly necessary optimization is to be accomplished by the server, which is both more difficult and more promising than leaving optimization to the applications.

Our proposed media server concept is based on a generalization of data independence called transformation independence. This abstraction reduces the creation, retrieval, and modification of media objects to what can be called the "pure application semantics". The consequence are multiple optimization dimensions being left for exploitation by the server. The VirtualMedia concept realizes transformation independence based on virtual media objects being described by filter graphs. With this concept, optimization can be basically characterized as the process of optimally matching transformation request graphs and materialization graphs.

Since we are just at the beginning of developing these concepts, there are still a lot of challenges to be mastered, like formalizing and evaluating the graph transformation algorithms and realizing the concept using available DBMS-technology and media server components.

Presented at: Seminar 99351 "Multimedia Database Support for Digital Libraries", Schloss Dagstuhl, Sept. 1^st, 1999.