An efficient and parallel file defragmentation scheme for flash-based SSDs
Guangyu Zhu, Jeongeun Lee, Yongseok Son · Proceedings of the 37th ACM/SIGAPP Symposium on Applied Computing · 2022
File fragmentation can increase I/O latency and decrease I/O throughput because reading a fragmented file can be followed by reading several scattered blocks. Thus, a defragmentation process is required to provide better I/O performance. However, an existing defragmentation process can be time-consuming on SSDs since it is based on HDD without consideration of SSD characteristics. For example, the process is performed by a single defragger in a serialized manner. To accelerate this defragmentation process, we propose an efficient and parallel file defragmentation scheme for flash-based SSDs. Our scheme exploits the internal parallelism of SSDs to reduce the execution time of the defragmentation process. To do this, we devise multiple defraggers to perform defragmentation and I/O operations for several files in parallel. We implement our scheme and evaluate it on a real machine with a flash-based SSD. The evaluation results show that the proposed scheme can reduce the execution time by up to 2.96x compared with an existing defragmentation tool, e4defrag.