The tremendous power and low price of today's computer systems have created the opportunity for exciting applications rich with graphics, audio and video. These new applications promise to support and even enhance the work we do in teams by allowing users to collaborate across both time and space. Despite their exciting potential, building distributed collaborative multimedia applications is very difficult and predicting their performance can be even more difficult. Performance predictions must deal with future hardware, future software, and future users and their requirements. The rate of change in new technologies continues to accelerate, with each new generation of hardware and software introducing new capabilities. Furthermore, distributed environments have a wider variety of potential configurations than do centralized systems.

To determine the computer resources needed to meet distributed application demands we have developed a flexible model and a method for applying it that allows us to predict multimedia application performance from the user's perspective. Our model takes into account the components fundamental to multimedia applications: latency, jitter and data loss. The contributions of this thesis, to both computer scientists and computer system developers, are: 1) a multimedia application quality model and method for predicting application performance and evaluating system design tradeoffs; 2) detailed performance predictions for three distributed collaborative multimedia applications: an audioconference, a ``flying'' interface to a 3D scientific database and a collaborative flight simulator; 3) the effects of system improvements on these multimedia applications; 4) a demonstration of measures of jitter that have been used by jitter researchers showing all reasonable choices of jitter metrics are statistically similar; and 5) experiment-based studies of the effects of high-performance processors, real-time priorities and high-speed networks on jitter.

In predicting the bottlenecks in quality for the above three applications, we have identified three general traits: 1) processors are the bottleneck in performance for many multimedia applications; 2) networks with more bandwidth often do not increase the quality of multimedia applications; and 3) performance for many multimedia applications can be improved greatly by shifting capacity demand from computer system components that are heavily loaded to those that are more lightly loaded.

Download:

• Full Document: ps, pdf