Consumption-Based Buffer Management for Maximizing System Throughputs of News-on-Demand Multimedia Systems

In a news-on-demand multimedia server, multiple multimedia streams are generally serviced in a cyclic fashion. Due to non-uniform playback rates and asynchronous arrivals of queries, there tends to be spare disk bandwidth in each service cycle. In this paper, we study the problem of dynamically using spare disk bandwidth and buffer to maximize the system throughput of a multimedia server. We introduce the concept of minimizing buffer consumption as the criterion to select an appropriate media stream to use the spare system resources. Buffer consumption measures not only the amount of buffer but also the amount of time such buffer space is occupied (i.e., the space-time product). Different alternatives to utilizing spare disk bandwidth are examined, including different rate-adjustable retrievals of an already activated stream and prefetching the next waiting stream. For rate-adjustable retrievals, we study buffer consumption-based and remaining time-based criteria for selecting an active stream to increase retrievals. Simulations are conducted to evaluate and compare different cases. The results show that (1) minimizng buffer consumption is the right criterion for maximizing the system throughput with spare disk bandwidth; (2) with sufficient buffer space or proper implementation, rate-adjustable retrievals of active streams perform better with minimal buffer consumption as a selection criterion than with minimal remaining time; (3) in general, prefetching a waiting stream incurs more buffer consumption, and thus is less effective than rate-adjustable retrievals of active streams in maximizing the system throughput.