SYNOPSISA high-performance file system designed for clusters.
DESCRIPTIONLustre® is a high-performance, massively-scalable, POSIX-compliant network file system designed for the world's largest high-performance compute clusters. Lustre is under active development from Sun Microsystems (http://www.lustre.org).
Lustre filesystems are made up of multiple services typically distributed across multiple nodes. Lustre clients can contact these server nodes over multiple high-speed network fabrics via LNET, the Lustre NETworking system. Clients then present the filesystem at a mount point to userspace.
A filesystem is comprised of a MDT , MetaData Target, which stores directory and file meta-information such as file ownership, timestamps, access permissions, etc, and a series of OSTs , Object Storage Targets, which hold the file data in one or more objects. There is typically not a 1:1 mapping of OST objects and what is presented as a file on a Lustre client.
Lustre and LNET are implemented as a series of kernel modules, for both servers and clients. LNET networks are defined in the modprobe.conf file on all nodes. Lustre is started on the clients and servers using the mount(8) command.
- Format a physical disk for use as a Lustre server's backend storage (aka target).
- Modify configuration information on a Lustre target disk.
- A helper program for mount(8) that starts Lustre servers and clients mounts the client filesystem.
- A low-level interface to control various aspects of Lustre
- A user-level interface to control Lustre-specific information for individual files. lustre_config.sh Format multiple Lustre targets simultaneously from definitions in a CSV file.