Lustre is an open-source parallel distributed file system used in the world of HPC. At last count, 60% of the Top 100 supercomputers use it as a file system. It was developed at Carnegie Mellon University more than eighteen years ago. The key features are the 1) Metadata Servers “MDS” 2) Metadata Targets “MDT” 3) Object Storage Servers “OSS” 4) Object Server Targets “MDT” and 5) Lustre clients. 

Like XFS and OpenZFS, Lustre file system supports petabytes of storage, hundreds of GB/s of I/O bandwidth, and scales to thousands of nodes. The way these file systems work, the more nodes there are, the better they perform because the workloads can be spread across several systems in a parallel state. The features include the following:    

  • More than 50% of the top 100 supercomputers use it
  • POSIX-compliant
  • Copy-on-write
  • Supports stripes, mirrors, RAID 1, 2, 5, and 6
  • Runs on commodity hardware
  • Deploys client-server architecture
  • Implemented entirely in the kernel as loadable modules
  • 5 Components: MDS, MDT, OSS, MDT, and clients
  • Metadata servers manage all metadata
  • OSS writes data to persistent storage
Scroll to Top