Abstract
A hybrid solid-state drive (SSD) consisting of both single-level-cell (SLC) and multi-level-cell (MLC) flash chips achieves a response time as fast as an SLC-flash-based SSD while maintaining the price of an MLC-flash-based SSD. It is supported by a software layer called flash translation layer (FTL) that contains an algorithm to efficiently store hot and cold data in the SLC- and MLC-flash chips, respectively. Unfortunately, previous FTLs for hybrid SSDs depended hot data identification solely on the write commands’ request size and reused former address mapping algorithms for managing the SLC- and MLC-flash chips. To address this limitation, we propose a “data pattern aware FTL (DPA-FTL)” algorithm. DPA-FTL enhances the hot data identification process by considering two characteristics of hot data: frequent update and irregular allocation. Furthermore, it compares former address mapping algorithms and selects a more appropriate algorithm for the hybrid SSD. According to our performance evaluation, DPA-FTL reduces the overall number of write and erase operations and reduces the deviation of erase operations.
| Original language | English |
|---|---|
| Pages (from-to) | 101-127 |
| Number of pages | 27 |
| Journal | Design Automation for Embedded Systems |
| Volume | 19 |
| Issue number | 1-2 |
| DOIs | |
| State | Published - 29 Mar 2015 |
Keywords
- Flash memory
- FTL
- Hybrid SSD
- Middleware