The Fundamental Limit of Flash Random Write Performance: Understanding, Analysis and Performance Modelling

The understanding, analysis and modelling of the fundamental limit of the sustained random write performance and endurance of Flash solid state drives (SSDs) are critical for Flash SSD vendors and storage system designers and practitioners. This not only helps design high-performance Flash SSDs, but also dictates how Flash can be integrated into today’s memory and storage hierarchy.
This paper analyzes the fundamental limit of the sustained random write performance of Flash SSDs. An empirical model is developed to compute the write amplification and performance slowdown factor for the greedy garbage collection policy under a pure random write workload. Potential causes of the commonly observed performance slowdown are investigated, and remedies suggested. In particular, we quantitatively demonstrate the potential for enhancing the random write performance and endurance by separating long-lived data from short-lived data inside Flash SSDs. Moreover, our theoretical results suggest a tiered storage system in which cold (i.e., infrequently used) data blocks are moved to a hard disk drive (HDD) to improve cost effectiveness and to reduce Flash memory utilization, which improves garbage collection performance.