DTFS: Exploiting the similarity of data versions to design a write-efficient file system in phase-change memory

Phase-change memory (PCM) has emerged as one of the most promising technologies to incorporate into the storage hierarchy of future computer systems. However, PCM has critical weaknesses in performing write operations. Specifically, the access latency and the energy consumption that occur during a write operation are about 6-10 times larger than those of a read operation in PCM. To cope with this situation, we analyze the write access characteristics of file systems, and observe that a large percentage of file write I/Os incurs small changes from the previous version. Based on this observation, we design a novel file system for PCM called DTFS (dual-tree file system) that incurs minimal writes to PCM. DTFS performs better than legacy file systems, but its reliability is as high as any file systems based on journaling or copy-on-write. Experiments with various workload conditions show that DTFS improves the file system performance and the energy consumption by 47% and 67% on average, respectively, compared to copy-on-write file systems such as ZFS and BtrFS.