release of the clustered file system , in most (~60%) Linux clusters, containing tens of thousands of nodes. Scalability in such large systems is achieved through a multi-component architecture. Key components of Lustre include metadata processing and storage servers (MDS), management servers (MGS), object storage servers (OSS), object storage targets (OST, supporting operation over ext4 and ZFS), and clients.
Key :
- persistent client cache (Persistent Client Cache), which allows the use of local storage such as NVMe or NVRAM as part of the global namespace of the filesystem. Clients can cache data related to created or existing files in the locally mounted filesystem with a cache (for example, in ext4). During the operation of the current client, these files are processed locally at the speed of the local filesystem, but, in case another client attempts to access them, they automatically migrate to the global filesystem.
- In routers automatic route discovery when using multi-path routing through different network interfaces (Multi-Rail Routing) and increased reliability of configurations with nodes that have multiple network interfaces.
- the 'overstriping' mode, in which multiple copies of stripe blocks for a single file can be contained within one object storage target (OST), allowing several clients to simultaneously perform write operations on the file without waiting for the lock to be freed.
- Now includes self-extending file placement schemes (Self-Extending Layouts), which enhance the flexibility of applying the PFL (Progressive File Layouts) mode in heterogeneous filesystems. For instance, when the filesystem includes small pools of storage based on fast Flash drives and large disk pools, the proposed capability allows for writing primarily to fast storage, and then automatically switching to slower disk pools once space runs out.
Source: opennet.ru
