Throughput optimization in disk-based real-time application specific systems

TL;DR: This work develops efficient optimization algorithms for data assignment on a single and multiple disks so that the seek time overhead is minimized and the importance of the synthesis problem and effectiveness of the proposed optimization algorithms are clear.

Abstract: Traditionally, application specific computations have been focusing on numerically intensive data manipulation. Modern communications and DSP applications, such as WWW, and video-on-demand, however, are increasingly more data management oriented. At the same time, the technological trends indicate that magnetic disk system performance is rapidly becoming a principal component in overall system behavior. We initiate system level research in synthesis of disk-based application specific system by introducing cost and throughput optimization problems for synthesis of disk- based application specific systems. We formulate, establish computational complexity, and develop efficient optimization algorithms for data assignment on a single and multiple disks so that the seek time overhead is minimized. Extensive experimental results clearly indicate the importance of the synthesis problem and effectiveness of the proposed optimization algorithms.