Delay-sensitive data retrieval refers to situations in which digitized information is fetched and converted for the presentation of audio, video, or some similar medium. Delays in the process would cause effects such as pops and clicks in audio and jitter in video. It would not merely be slow or have poor performance but, as the authors say, it would have glitches of a “very annoying character.” At present, the most typical case of handling of delay-sensitive media is a single workstation accessing a file in local storage. However, there is research and commercial activity aimed at network file servers that would deliver multimedia content over a network for multiple users. The paper presents a theoretical framework for multiple simultaneous delay-sensitive file retrieval. The authors address the strictest form of the problem, tolerating no missed deadlines as data are retrieved to be transmitted over multiple channels and making no assumptions about the data. The discussion is in terms of retrieval and digital audio, but could easily apply to storage (recording) and video.

The paper begins with a review of previous work. It sets out the requirements for real-time playback of audio in terms of reading samples from various locations on a disk to be consumed by processes that convert the data to analog sound. The authors establish necessary and sufficient bounds on buffer space. They define an algorithm with a game and define-and-compare strategies for the disk head movement. They point out that their approach, the sorting-set algorithm, is fundamentally equivalent to the grouped sweeping scheme developed concurrently by Yu, Chen, and Kandlur [1]. The paper closes with a method of estimating performance for a set of situations and analysis of previous work.

The paper is clearly written and suitable for professionals in the field and attentive students. The authors appear to know their field and provide motivation for the theoretical work.