Multi-disk Scheduling for High-Performance RAID-0 Devices

摘要

High-Performance I/O subsystems have become a must for multimedia systems, such as video servers [15,16,18]. The proposing of the Intelligent I/O (I2O) specifications [11] provides hardware vendors an operating-system-independent architecture in building their solutions for high-performance I/O subsystems. This paper targets one of the most important performance issues in building an I2O RAID-0 device, which is an important I2O implementation. We explore multi-disk scheduling for I2O requests, which are usually associated with soft deadlines to enforce quality-of-service requirements. The idea of Least-Remaining-Request-Size-First (LRSF) is proposed for the request-level multi-disk scheduling with the objective to improve the response-time requirements of I/O sub-systems. The proposed scheduling algorithm is then extended to resolve the starvation problem and for SCAN-like disk services. Finally, we exploit pre-fetching for I2O RAID-0 devices to further improve their performance. The proposed methodologies are verified by a series of experiments using realistic and randomly generated workloads.