OpenZFS Introduces Sync Parallelism for Enhanced Write Performance Scalability

OpenZFS, the popular file system for Linux servers and home labs, has received a significant performance enhancement. The new feature improves write performance scalability by allowing parallel syncing of multiple datasets/objects and reducing locking contention.

This optimization was developed by Brian Behlendorf from the Lawrence Livermore National Laboratory (LLNL) over the summer. Tests conducted on an AMD EPYC server running OpenZFS showed a remarkable improvement in write bandwidth. The write performance increased from 1328MB/s to 2495MB/s, as reported by the developer. Another developer also observed a significant boost in FIO write bandwidth, jumping from 1386MB/s to 2449MB/s.

In addition to the substantial increase in write performance, the new feature also reduces locking contention. This improvement further enhances the overall efficiency of OpenZFS.

The ZFS sync parallelism for OpenZFS has been merged into the Git repository, as announced in the recent pull request. Detailed information about this enhancement can be found in the pull request link provided.

This update is a significant milestone for OpenZFS users, as it greatly enhances write performance scalability and improves the overall efficiency of the file system. The optimization is expected to benefit users in various scenarios, from Linux servers to home labs and DevOps environments.

Source: Phoronix.