SMART based Dependability Analysis Methodology for SSD: Measurements and Implications

These days, the capacity of flash memory is increasing rapidly by applying MLC (Multi Level Cells) techniques and raising the storage density per unit area. However, such technologies cause the interference among cells such as read and write disturbances and deteriorate the program/erase cycles and data retention problems, leading to the dependability as the primary concern of flash memory based storage devices. As a starting point to mitigate the concern, in this paper, we propose a new methodology and tool that can explore the dependability issues of SSD (Solid State Drives) systematically and quantitatively. Using the SMART (Self Monitoring Analysis and Reporting Technology), it can measure diverse attributes such as the average erase count of blocks, I/O response time, and the number of bad blocks under the various workloads. Also, it analyzes the measurements and provides their implications including the expected lifetime of SSD, the effects of sequential/random workloads on endurance, and the effects of bad blocks on performance. Real experiments with six commercial SSDs have shown that the proposed methodology can interpret the internal behaviors appropriately and suggest useful information to evaluate and enhance the dependability of SSDs.