솔리드 스테이트 드라이브에서 데이터 신뢰성과 성능 향상을 위한 소프트웨어적 접근

Abstract

Ever since its first introduction in 1984, NAND flash memory is widely used in various storage devices as the storage medium. Especially, Solid State Drive (SSD) which was composed of multiple NAND flash became better to meet the levels required by computer system in terms of performance, capacity and costs. However, PCIe based SSD requires more complicated hardware/software techniques (ex. write buffering technique, write reduction technique, parallel architecture) because PCIe provides more bandwidth than SATA/SAS interface. If PCIe based SSD used all of these techniques to satisfy requirements for computer system, it might cause the problem that it can’t guarantee the reliability in storage device in case of sudden power-off. Therefore this dissertation proposed Dual-Region Write Buffering(DRWB), Metadata Embedded Write (MEW), and zero-free Flash Translation Layer (zf-FTL), which guarantee the data reliability using software approach with minimizing the capacity of supercapacitor while improving performance. The Dual-Region Write Buffering (DRWB) is a novel write buffer scheme that implements logically non-volatile write buffer using large sized DRAM and small capacity capacitor. Experimental results show that the proposed DRWB scheme enables us to achieve the same effect with the NVRAM write buffer, in terms of the data reliability, without noticeable performance degradation. And MEW scheme exploits the compression-induced internal fragmentation in the data area to store rich metadata for fast and complete recovery. We performed extensive experiments to examine the performance of MEW. The performance overhead of MEW is 3% in the worst case, in terms of the write amplification factor, compared to the pure compression-based FTL that does not have any recovery scheme. Finally, zero-free FTL (zf-FTL) effectively reduces the amount of data that will be actually written to the flash memory without requiring any additional data structure or dedicated hardware. Experimental results show that the zf-FTL outperforms general page-mapped FTL in terms of both the number of page read/write operations and the average I/O latency.