The authors explain some of the critical issues involved in the manipulation of delay-sensitive data, such as a digitized audio sample, and then go on to establish a theoretical framework for analyzing the performance of playback. Delay-sensitive data have different requirements from other real-time data: for example, a critical requirement is that there be no delays in the middle of the process of retrieving a sample, decompressing it, and converting it to analog sound, and this requirement may be achieved at the cost of an initial delay and the use of buffers. The paper includes a review of the literature; definitions, theorems, and proofs; consideration of factors such as seeks, buffers, single- and multi-channel playback, and storage placement; and a report on a simulation exercise testing an algorithm for buffer allocation.

This lengthy paper is well written, and the graphs, formulas, and proofs are clearly presented. It is written at an appropriate level for researchers in the field. The work itself appears sound and well-grounded in issues in the field. A deficit of the work, which may be difficult to address, is that it is purely theoretical. Commercial and research systems already exist that perform retrieval and utilization (the authors would use the term “consumption”) of digital audio and other delay-sensitive data. It would have been interesting to read how the authors’ theoretical framework corresponds with actual cases. Certain factors emphasized in the theoretical model, such as buffering and blocking, may require different treatment because of the specific approaches taken in a client/server networking situation or a situation involving workstations with special processors. The commercial systems may make use of proprietary algorithms not available for analysis, however.