Vendor-neutral by design
Support options for Lustre have historically followed the storage appliance. We sell engineering, not arrays. The code stays GPLv2.
Independent Lustre stewardship
The parallel file system behind the world’s largest machines.
The Lustre Collective
Vendor-neutral development, support, and performance engineering for Lustre — the open-source filesystem that still runs the majority of Top100 HPC systems and the largest AI training clusters.
60%+of Top100 HPC systems run Lustre, a lead it has held for more than a decade.
10+ TB/sproduction bandwidth, with 100M+ IOPS and 700 PB+ in a single namespace today.
2.17 GACommunity release, December 2025. LTS 2.15.8 ships in parallel. 2.18 targets Q4 2026.
Founded 2025Independent, founder-owned, no venture capital — structured to last like a foundation.
Why TLC exists
Lustre is free. Running it at the limit is not.
The Lustre Collective launched at SC25 in November 2025. Three original Cluster File Systems veterans — Peter Jones, Andreas Dilger, and Colin Faber — formed an independent company so sites can get Lustre expertise that is not tied to a hardware vendor, a trademark owner, or a VC exit.
The people who write it
Andreas Dilger has led Lustre architecture since 1999. The rest of the founding team has shipped community releases, appliances, and exascale deployments for two decades.
Built for HPC and AI
Checkpoint storms, GPU-direct paths, multi-tenant clouds, and 100k-GPU training clusters are the workload, not a stretch goal.
On the floor
Where Lustre actually runs
As of the June 2026 TOP500, Lustre remains the campaign filesystem on the U.S. DOE exascale systems and the capacity tier beside DAOS on Aurora. It is also the storage under flagship AI machines and first-party cloud Lustre services.
Exascale HPC
- El Capitan (LLNL) — #2 on TOP500 June 2026, 1.809 EF, Lustre campaign storage.
- Frontier (ORNL) — first verified exascale system, still among the largest operational Lustre deployments on earth (>700 PB class).
- Aurora (Argonne) — DAOS for the flash campaign tier, with the 100 PB Flare Lustre system for center-wide data.
- LUMI (CSC / EuroHPC) — large Lustre capacity tier alongside flash and object storage.
AI factories and clouds
- xAI Colossus — Lustre-based primary storage for one of the largest GPU training clusters built.
- NVIDIA EOS / DGX SuperPOD — reference architectures that specify Lustre-class parallel storage.
- AWS FSx for Lustre, Azure Managed Lustre, and Google Cloud Managed Lustre (GA July 2025) — first-party managed Lustre in every major cloud.
Current tree
Lustre 2.17 is out. 2.18 is on the clock.
2.17.0 — GA 29 December 2025
- Hybrid I/O — automatic buffered vs direct I/O for large transfers (LU-18033).
- Dynamic LNet NID configuration — ephemeral interfaces without bouncing the config (LU-18815).
- Nodemap enhancements — finer multi-tenancy, filesets, and root isolation (LU-17431).
- Servers on RHEL 9.7; clients on RHEL 9.7 / 10.1, SLES 15 SP7, Ubuntu 24.04.
2.18 — targeted Q4 2026
- File-level erasure coding (read-only FLR-EC) with ORNL, Whamcloud, and TLC.
- Fault-tolerant MGS — replicated config and recovery across MDS nodes.
- Client data compression and RHEL 10 server support.
- Roadmap scale: 200M+ IOPS, 10M+ metadata ops/s, 500B+ files, 1 EB-class namespaces.
LTS 2.15.8 shipped 2 December 2025. Community downloads live at lustre.org/download.
Need the filesystem that already won, with a team that still ships it?
Production support, architecture reviews, custom features, and training — from an independent company whose only product is Lustre.
Talk to a Lustre expert