Improving ZFS Synchronous Write, O_SYNC|O_DSYNC Performance (Doc ID 1347387.1)

Last updated on OCTOBER 22, 2024

Applies to:

Goal

A number of workloads, including the Oracle Database, NFS and applications calling fsync(3C) or fdsync require synchronous semantics. When an application opens a file with the O_SYNC|OD_SYNC flag, write(2) system calls to that file return only after the data is committed to the stable storage. This behavior is different from buffered writes which return after writing the data to the file system cache to be later committed to the stable storage (5 seconds or less in ZFS)

ZFS uses the ZIL (ZFS Intent Log) to process synchronous I/O.  ZFS offers several tunables to improve write performance.

Solution

My Oracle Support provides customers with access to over a million knowledge articles and a vibrant support community of peers and Oracle experts.