Pre-emptive but safe interval caching for real-time multimedia systems

The fundamental problem in developing high-performance multimedia database servers is that images, audio and video data differ from numeric or text data in their characteristics, and hence require different handling techniques. Digital audio and video streams consist of a sequence of media quanta (video frames or audio samples) which convey meaning only when presented contiguously on time. In order to ensure continuous delivery of video or audio, reservation of various resources (e.g., disk bandwidth, memory, network bandwidth, etc.) is required before starting playback of a stream. The maximum bandwidth that can be reserved on each disk will depend on the number of concurrent streams it can support with an acceptable level of jitter or hiccup, i.e., without violating on-time delivery constraints. Read-ahead buffering, where blocks are read and buffered ahead of the time they are needed, has been introduced. Although early work on continuous media storage systems assumed separate buffers for each client, more recent work [11, 15] has assumed a global buffer cache, similar to buffer caches in traditional storage systems [10, 20] where the cached data can be shared among all the clients. The rationale for using a buffer cache for multimedia database systems is the same as that for traditional storage systems: namely, the reduction of disk I/O. Unlike in traditional systems, LRU does not yield a high cache hit ratio for multimedia data access [3]. In [6, 8, 16], a caching scheme for a video server has been proposed. The scheme exploits temporal locality accessing the same multimedia object, by caching intervals between successive streams (preceding stream and following stream), and is referred to as interval caching. If none of the cached intervals are replaced or preempted until they have finished their playback, the following streams of those intervals experience no hiccup without access to the disks. However, since the interval caching allows cache replacement, it cannot guarantee hiccup-free playback as in [17] if additional streams are admitted using the saved disk bandwidth. Ozden et al. proposed several schemes which improve the cache hit ratio in multimedia storage systems [17]. However, their cache replacement schemes cannot guarantee the hic-