ParaFS: a log-structured file system to exploit the internal parallelism of flash devices

Abstract

File system designs are undergoing rapid evolution to exploit the potentials of flash memory. However, the internal parallelism, a key feature of flash devices, is hard to be leveraged in the file system level, due to the semantic gap caused by the flash translation layer (FTL). We observe that even flash-optimized file systems have serious garbage collection problems, which lead to significant performance degradation, for write-intensive workloads on multi-channel flash devices. In this paper, we propose ParaFS to exploit the internal parallelism while ensuring efficient garbage collection. ParaFS is a logstructured file system over a simplified blocklevel FTL that exposes the physical layout. With the knowledge of device information, ParaFS first proposes 2-D data allocation, to maintain the hot/cold data grouping in flash memory while exploiting channel-level parallelism. ParaFS then coordinates the garbage collection in both FS and FTL levels, to make garbage collection more efficient. In addition, ParaFS schedules read/write/erase requests over multiple channels to achieve consistent performance. Evaluations show that ParaFS effectively improves system performance for write-intensive workloads by 1.6× to 3.1×, compared to the flash-optimized F2FS file system.