ossfs 2.0

Performance improvements

ossfs 2.0 offers significant performance improvements over ossfs 1.0 in sequential read and write operations, and in high-concurrency reads of small objects. For more information about the performance of ossfs 2.0, see Performance testing.

• Sequential write performance: In single-threaded, large-object sequential write scenarios, ossfs 2.0 provides nearly 18 times the bandwidth of ossfs 1.0.

• Sequential read performance: In single-threaded or multi-threaded (4 threads) large-object sequential read scenarios, ossfs 2.0 provides more than 3 times the bandwidth of ossfs 1.0.

• Concurrent small-object read performance: In high-concurrency (128 threads) small-object read scenarios, ossfs 2.0 provides more than 20 times the bandwidth of ossfs 1.0.

Use cases

ossfs 2.0 is suitable for scenarios that require high-performance storage access, such as AI training, inference, big data processing, autonomous driving, and other compute-intensive workloads. These workloads primarily involve sequential and random reads, sequential (append-only) writes, and do not require full POSIX semantics.

Runtime environment

ossfs 2.0 is built on Filesystem in Userspace (FUSE).

Limitations

• Permission requirements: The AccessKey you use must have full permissions for the target bucket or prefix. Insufficient permissions may cause the mount operation to fail or other operations to behave unexpectedly.

• Storage class limitations: We recommend against mounting buckets in the Archive Storage, Cold Archive Storage, or Deep Cold Archive Storage classes.

• Object and directory name limitations: Object and directory names are limited to 255 characters due to a Linux constraint. Objects or directories in OSS with names that exceed this limit are not visible in the mount point.

• Object read limitations: ossfs uploads newly written objects to OSS only after the corresponding file handles are closed. Reading an object before its handle is closed may cause an error.

• Object write limitations: Random writes and concurrent writes to the same object are not supported. The default part size is 8,388,608 bytes (8 MiB), which limits the maximum size of a writable object to 83,886,080,000 bytes (78.125 GiB). You can configure the part size by using the upload_buffer_size mount option.

• Object rename operations (non-atomic)

  • Objects: The operation first copies the source object to the destination and then deletes the source object.

  • Directories: The operation copies all objects under the source directory to the destination and then performs a bulk deletion of the source objects. By default, the rename operation is limited to 2 million descendant objects. You can change this limit by using the rename_dir_limit mount option.

• Concurrent write consistency: ossfs 2.0 does not guarantee data consistency when multiple clients mount the same bucket and write to the same object concurrently.

• POSIX API compatibility: Partially compatible. For details, see POSIX API support.

Features

References

• Install ossfs 2.0

• Configure ossfs 2.0

• Mount a bucket using ossfs 2.0

• ossfs 2.0 mount options

• ossfs 2.0 performance tests

• Object and directory recognition mechanism

• Reduce OSS requests and improve mount point performance

• FAQ

• Release history