To prevent the overhead of creating new metadata structures for every small write, SnapFiles employs a . Writes are buffered, chunked, and hashed in memory. The metadata is only flushed to the Metastore when a file handle is closed or a sync is issued. This batch-processing approach significantly reduces IOPS (Input/Output Operations Per Second) overhead on the Metastore.
In sequential write tests, SnapFiles showed a 5% overhead compared to raw NFS due to the computational cost of hashing. However, in write-heavy environments with duplicate data (e.g., VM image storage), SnapFiles achieved due to deduplication. snapfiles
SnapFiles: A High-Performance Distributed Filesystem for Immutable Atomic Snapshots To prevent the overhead of creating new metadata