Parts/Storage/Lustre
Storage · Lustre · 2003
Lustre
The parallel file system underneath most large HPC deployments, shipped by several vendors as an appliance rather than sold by one.
Why this part
Lustre is the reason most large machines can be read from at all, and it is owned by nobody: it began at Carnegie Mellon, was funded through the ASCI programme, passed through Cluster File Systems to Sun, then to Oracle, and when Oracle abandoned it the community forked it and carried on. It is open source, so the vendors on this site sell appliances built from it rather than the filesystem itself. Its architecture separates metadata from data, which is what lets thousands of clients write in parallel and also what makes a metadata server the thing that falls over first. Nearly every storage part in this taxonomy is either Lustre in a box or a deliberate attempt to replace it.
Systems using this part
- Systems
- 25
- Rmax underneath
- 294.9 PFlop/s
- Units recorded
- -
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| Arrhenius 2026-06 | Storage | Sweden | - | 66.82 PFlop/s | Hewlett Packard Enterprise | #42 → |
| Venado 2024-04 | Storage | USA | - | 98.51 PFlop/s | Hewlett Packard Enterprise | #26 → |
| Vega 2021-04 | Storage | Slovenia | - | 6.92 PFlop/s | DDN | #106 → |
| HiPerGator AI 2021 | Storage | USA | - | 17.49 PFlop/s | DDN | #106 → |
| Phoenix (PACE) 2020 | Storage | USA | - | 1.84 PFlop/s | DDN | #277 → |
| Mustang (AFRL) 2018-12 | Storage | USA | - | 3.22 PFlop/s | DDN | #48 → |
| TSUBAME3.0 2017-08 | Storage | Japan | - | 8.13 PFlop/s | DDN | #59 → |
| Trinity 2017-06 | Storage | USA | - | 20.16 PFlop/s | Cray now Hewlett Packard Enterprise | #7 → |
| Piz Daint 2017-06 | Storage | Switzerland | - | 21.23 PFlop/s | Cray now Hewlett Packard Enterprise | #68 → |
| JAIST Cray XC40 2017 | Storage | Japan | - | 584.77 TFlop/s | Cray now Hewlett Packard Enterprise | #339 → |
| Cori 2016-11 | Storage | USA | - | 14.02 PFlop/s | Cray now Hewlett Packard Enterprise | #5 → |
| Reedbush-U 2016 | Storage | Japan | - | 459.83 TFlop/s | Hewlett Packard Enterprise | #362 → |
| JAEA SGI ICE X 2015-11 | Storage | Japan | - | 1.93 PFlop/s | SGI now Hewlett Packard Enterprise | #35 → |
| Shaheen II 2015-03 | Storage | Saudi Arabia | - | 5.54 PFlop/s | Cray now Hewlett Packard Enterprise | #7 → |
| Spirit (AFRL) 2013-06 | Storage | USA | - | 1.42 PFlop/s | SGI now Hewlett Packard Enterprise | #15 → |
| Blue Waters 2013-03 | Storage | USA | - | - | Cray now Hewlett Packard Enterprise | - |
| Titan 2012-11 | Storage | USA | - | 17.59 PFlop/s | Cray now Hewlett Packard Enterprise | #1 → |
| CRIEPI SGI ICE X 2012-07 | Storage | Japan | - | 582.10 TFlop/s | SGI now Hewlett Packard Enterprise | #39 → |
| RWTH Compute Cluster 2011 2011 | Storage | Germany | - | 219.84 TFlop/s | Bull now Atos | #32 → |
| Jaguar 2009-11 | Storage | USA | - | 1.76 PFlop/s | Cray now Hewlett Packard Enterprise | #1 → |
| Pleiades 2008-11 | Storage | USA | - | 5.95 PFlop/s | SGI now Hewlett Packard Enterprise | #3 → |
| Rostand 2008 | Storage | France | - | 106.10 TFlop/s | Sun Microsystems now Oracle | #11 → |
| BlueGene/L 2004-11 | Storage | USA | - | 478.20 TFlop/s | IBM | #1 → |
| FASTER (Texas A&M) | Storage | USA | - | - | DDN | - |
| Bluecrab | Storage | USA | - | - | Intel | - |
“Units recorded” sums only the edges where a public source states a quantity. It is a floor, not a total, and should never be read as installed base.