Lustre is a scale-out file system typically used in High Performance Computing applications. As a parallel distributed file system, Lustre is deployed where high parallelism and bandwidth are required. Lustre is open source software developed originally as a research project at Carnegie Mellon and since has had commercial implementations distributed and supported as complete storage solutions. The first such company with a distribution and product was ClusterStor which was subsequently acquired by Sun Microsystems. Through different acquisitions and variations, the distribution and support for the commercial version is now with DataDirect Networks with Lustre offered in the DDN Exascaler product. The open source distribution continues and is freely downloadable. Many add-on enhancements were developed independently and may be installed on top of the Lustre environment.
Architecturally there is no limit to the number of nodes and capacity but actual implementations have more modest with the largest containing 2 Billion files and 55 PB in a production environment. With the open source offering and high performance, a majority of Lustre implementations are in research labs and universities – often attached to supercomputers.
The key design element for Lustre is a separation of the data and metadata with metadata on MetaData Servers (MDS) and MetaData Target (MDT) devices and data stored on Object Store Targets (OST), not to be confused with object storage systems. Sun added to the Lustre architecture the alternative to have a ZFS storage system as the target for stored data. ZFS adds greater data protection, replication, and data reduction to Lustre whereas the default storage “backend” lacked these capabilities. An implementation may have either a ZFS backend or standard storage devices.
A POSIX interface with a Lustre client may be used for access and exploiting parallelism. In addition, NFSv3 and CIFS/SMB (via SAMBA) are supported for standard file serving as a NAS system. The Lustre client (also called a Linux Client) also requires a Lustre filesystem device driver loaded into the kernel on a client system. A single, unified system (global namespace) is presented regardless of the number of nodes in the Lustre configuration.
Lustre software can be installed on different servers and utilize multiple storage systems that may be RAID arrays, JBOD devices, or ZFS systems.
Lustre Software Product Brief includes:
- Overview
- Highlights
- Usage
- EvaluScale Product Review Methodology
- Evaluator Group Opinion
Download now to read the full product brief!
