Bcachefs Tools 1.38.6 Released, Major Performance Enhancements Achieved

Kent Overstreet announced today the release of Bcachefs Tools 1.38.6, the userspace toolset for the Bcachefs file system. Bcachefs is a copy-on-write (COW) file system for Linux, and version 1.38.6 is a significant update featuring new functionalities along with extensive performance improvements throughout its codebase. This version ensures no on-disk format corruption, allowing existing users to upgrade safely.

The primary highlight of Bcachefs Tools 1.38.6 is its performance enhancements. Overstreet commented during the release, “The performance improvements span the majority of the codebase, with profiling done on numerous workloads and benchmarks.” Additionally, this version now supports up to 255 storage devices within a single file system. Furthermore, packages for Ubuntu 26.04 LTS are now available on apt.bcachefs.org, making testing more accessible.

Lock-Free Journaling

One of the most notable improvements in this release is the optimization of the journaling code. The update primarily targets lock contention, delivering remarkable performance gains, especially for O_SYNC and fsync workloads in multi-threaded environments.

Specifically, journal flushing has been made entirely lock-free, significantly boosting throughput in high-concurrency scenarios. The journal pin list mechanism has also been updated from sharing a main journal lock to having individual locks for each pin list. Additionally, journal pin FIFOs are now dynamically resizable at runtime, avoiding unnecessary throttling in workloads that generate numerous O_SYNC or fsync operations.

Improvements have also been made to journal reads, reducing the amount of journal data read by identifying the latest entries on each device through binary search. Leveraging these optimizations, the default journal size has also been increased.

Btree Iterator Optimization

The Btree iterator and transaction commit code, core components of Bcachefs, have undergone profiling and optimization. Notably, the usage of instruction cache (icache) has been significantly reduced. This improvement directly boosts performance in large-scale workloads that require efficient CPU cache line usage.

Btree Sharding Enhancements

Another critical update in this release is the enhancement of Btree sharding. Previously, new inode allocations were sharded based on the current CPU. In version 1.38.6, sharding is now based on process IDs (pids), reducing scenarios where load concentrates on specific CPUs.

Additionally, a mechanism has been introduced to ensure that Btree nodes do not cross shard boundaries. When lock contention occurs, the thread attempts to migrate to the CPU associated with the shard containing the data. These changes have led to a significant reduction in lock contention, particularly in dbench benchmarks with a high number of clients.

Improvements to Multi-Threaded Write Buffers

The flushing of Btree write buffers in multi-threaded environments has been improved as well. Lock inversion stalls have been reduced, enhancing stability in high-concurrency workloads. Moreover, runtime introspection capabilities have been significantly strengthened, allowing for more detailed monitoring of the system’s state during operation.

Future Benchmarking Prospects

Phoronix has indicated the possibility of conducting new comparative benchmarks for Linux file systems in the future. Given that Btrfs has made large folio support a default feature and considering other file system improvements introduced in Linux 7.2, comparisons featuring Bcachefs are expected to draw significant interest.

Although Bcachefs remains in an experimental status, its advantages in scalability and performance over alternatives like ZFS and Btrfs are becoming increasingly evident. This release, in particular, has improved performance in multi-device environments, expanding its potential applications in enterprise storage scenarios.

Editorial Opinion

This release highlights the maturity of Bcachefs as it evolves into a stable and practical file system. Notably, the reductions in lock contention and the enhancements in sharding could serve as key differentiators for modern storage server environments equipped with numerous NVMe SSDs. Over the coming months, we may see growing adoption of Bcachefs as a standard feature in major Linux distributions. However, caution is still advised for deploying it in production environments, especially due to the remaining challenges regarding disaster recovery and online maintenance functionality.

In the long term, this release could mark a turning point in the expansion of the Bcachefs ecosystem. The support for up to 255 devices offers a way to manage large-scale storage pools within a single layer, something that traditional file systems have struggled with. Should Bcachefs establish itself as a stable presence in the Linux kernel’s mainline, it could potentially replace some of the territory currently dominated by ZFS. However, the heavy reliance on Kent Overstreet as the sole developer remains a concern, and the development of a robust community maintenance framework is an urgent need.

References

• Phoronix — Published on 2026-06-17T20:06:22.000Z