Improving the I/O performance of real-time database systems with multiple-disk storage structures

摘要

Time-constrained I/O scheduling is one of the most critical issues in the design and implementation of real-time database systems (RTDBSs) because the I/O time usually dominates the total execution time of a transaction. The performance of single-disk storage systems may not be appropriate for RTDBSs due to the large discrepancy between the disk I/O access/waiting time and the processor/main memory processing time. This paper proposes a new disk system which uses a multiple-disk structure similar to a RAID (Redundant Array of Inexpensive Disks) to improve the I/O performance for RTDBSs. We also propose a new disk scheduling algorithm called ED-PT (Earliest Deadline first-Prior Test) for determining the feasibility of I/O requests. This algorithm gives a priori knowledge about whether an arriving I/O request can meet its deadline, making it possible to reject an infeasible request before it is executed and thus this avoids wasted disk access effort if late completion of requests is not allowed. Our experimental results show that the percentage of requests missing deadlines decreases dramatically in a multiple-disk system when compared to a single disk system.