Overcoming Workstation Scheduling Problems in a Real-Time Audio Tool

The recent interest in multimedia conferencing is a result of the incorporation of cheap audio and video hardware in today’s workstations, and also as a result of the development of a global infrastructure capable of supporting multimedia traffic the Mbone. Audio quality is impaired by packet loss and variable delay in the network, and by lack of support for real-time applications in today’s general purpose workstations. A considerable amount of research effort has focused on solving the network side of the problem by providing packet loss robustness techniques, and network conscious adaptive applications. Effort to solve the operating system induced problems has concentrated on kernel modifications. This paper presents an architecture for a real-time audio media agent that copes with the problems presented by the UNIX operating system at the application level. The mechanism produces a continuous audio signal, despite the variable allocation of processing time a real-time application is given under UNIX. Continuity of audio is ensured during scheduling hiccups by using the buffering capabilities of workstation audio devices drivers. Our solution also tries to restrict the amount of audio stored in the device buffers to a minimum, to reduce the perceived end-to-end delay of the audio signal. A comparison between the method presented here (adaptive cushion algorithm), and that used by all other audio tools shows substantial reductions in both the average end-to-end delay, and the audio sample loss caused by the operating system.