Dynamic Detection of Deterministic Disk Access Patterns

As components within a system increase in performance at different rates, the components with the slowest rate of increase become bottlenecks for the entire system. In personal computing devices, increases in performance of hard disk drives are far outstripped by improvement in processor and main memory speeds, and are therefore a good candidate for performance analysis and examination. Many strategies exist already for improving disk performance. In this work, we focus on detecting deterministic disk access patterns, or patterns of disk access that can be shown to repeat over time. Once found, several strategies for harnessing this determinism are discussed. We introduce a novel method for efficiently detecting and monitoring disk for deterministic patterns, and show that this tool can reveal that more than half of disk access traces exhibit determinism that can be harnessed for improved performance.