BlueStore was designed to address these limitations. Following the development of NewStore, it was obvious that trying to use a POSIX filesystem as the underlying storage layer in any approach would introduce a number of issues that were also present in filestore. In order for Ceph to be able to achieve a guaranteed level of performance, that was expected from the underlying storage, it also needed to have direct block-level access to the storage devices without the additional overheads of a separate Linux filesystem. By storing metadata in RocksDB and the actual object data directly on block devices, Ceph can leverage much better control over the underlying storage and at the same time provide better performance.