Lustre (file system)

Lustre (file system)
Name	Lustre
Developer	Sun Microsystems, Intel, Hewlett Packard Enterprise, OpenSFS, EOFS
Full name	Lustre
Introduction date	2003
Directory struct	B-tree
File struct	Extents
Max file size	8 EiB
Max volume size	512 EiB
Features	Parallel access, High availability, Data striping
Os	Linux
Website	https://www.lustre.org

Contents

Overview
Architecture
Features
Deployment and usage
History and development
See also

Lustre (file system). Lustre is a type of parallel distributed file system designed for large-scale cluster computing and supercomputing environments. It is renowned for providing high-performance access to petabyte-scale data storage across thousands of client nodes. The architecture separates metadata and object data services, enabling scalable performance for demanding workloads in fields like scientific computing and artificial intelligence.

Overview

Lustre is a key enabling technology for some of the world's most powerful supercomputers, including systems at Oak Ridge National Laboratory and Lawrence Livermore National Laboratory. It was originally developed through a collaboration between Sun Microsystems and the United States Department of Energy to meet the extreme I/O demands of high-performance computing (HPC). The file system's name is a portmanteau of "Linux" and "cluster," reflecting its core design principles. Its development and governance have been stewarded by organizations like OpenSFS and EOFS.

Architecture

The Lustre architecture is composed of several discrete, networked components that communicate via the Lustre Networking (LNet) protocol. At its core are Metadata Servers (MDS) that manage namespace operations and store file metadata on a dedicated Metadata target (MDT). Actual file data is stored on one or more Object Storage Servers (OSS), each managing multiple Object Storage Targets (OST). Client nodes run the Lustre client software, which presents a POSIX-compliant interface and interacts directly with the OSS nodes for data I/O, bypassing the MDS for most data operations. This separation is fundamental to achieving high aggregate bandwidth.

Features

A primary feature of Lustre is its support for data striping, which allows a single file to be distributed across multiple OSTs and OSS nodes, dramatically increasing I/O bandwidth. It provides strong data consistency through a distributed lock manager and supports advanced security features like integration with Kerberos. The system is designed for high availability through failover capabilities for MDS and OSS nodes, often managed by tools like Pacemaker (software). Lustre also supports quotas, snapshots, and progressive file system checking utilities.

Deployment and usage

Lustre is predominantly deployed in large-scale research and government facilities, such as the Texas Advanced Computing Center and the National Energy Research Scientific Computing Center. It is the file system of choice for many TOP500 systems, including former number-one systems like Titan (supercomputer) and Summit (supercomputer). Commercial entities in sectors like geophysical exploration and media and entertainment, such as Weta Digital, also utilize Lustre for managing massive datasets. Deployment typically involves specialized storage area network hardware from vendors like DDN Storage, Hewlett Packard Enterprise, and DataDirect Networks.

History and development

The Lustre project began in 1999 at Carnegie Mellon University as a research project before being acquired by Sun Microsystems in 2007. Following the acquisition of Sun by Oracle Corporation in 2010, development slowed, prompting the HPC community to form the Open Scalable File Systems (OpenSFS) consortium to fund critical development. In 2012, Intel acquired the Lustre assets from Oracle and later transferred them to the OpenSFS and EOFS (European Open File Systems) organizations, which now coordinate open-source development. Major version milestones include Lustre 2.0, which introduced significant ZFS integration, and ongoing work to enhance cloud computing integration and non-volatile memory support.