Fwrite: An Approach to Optimize Write Performance of ZFS on Linux

ZFS file system has been widely used as it provides a file system interface and a block device interface. ZFS ignores the workload features of sequential writes. Writes to a block device with an unaligned I/O pattern will cause inefficient "read-modify-write" operations. In this paper, we propose an approach, called Fwrite, to optimize write performance of ZFS on Linux. This approach merges sequential write requests into fewer and larger I/O requests, which reduces the number of reads performed by "read-modify-write" operations. Experimental results demonstrate that this approach improves the original ZFS on Linux by 41.1-65.5% in sequential writes.