Livermore Computing (LC) recently expanded the capacity and network bandwidth of three of its Lustre file systems—lscratchd, lscratch6, and lscratch7.
In the Open Computing Facility (OCF) Collaboration Zone (CZ), the /p/lscratchd file system capacity was expanded to 5.7 PB, and network bandwidth is now 90 GB/s. In the Secure Computing Facility (SCF), both /p/lscratch6 and /p/lscratch7 capacities were expanded to 5.7 PB, and network bandwidth is now 90 GB/s.
The following tables summarize the current capacity and network bandwidth of each Lustre file system and maximum Lustre bandwidth between each file system and computing system.
| OCF File System | Bandwidth (GB/s) | Capacity (PB) |
|---|---|---|
| lscratchrzb | 18 | 1.2 |
| lscratchd | 90 | 5.7 |
| lscratche | 90 | 5.7 |
| lscratchf | 36 | 2.4 |
| lscratchv | 106 | 6.7 |
| SCF File System | Bandwidth (GB/s) | Capacity (PB) |
|---|---|---|
| lscratch1 | 850 | 53 |
| lscratch3 | 90 | 5.7 |
| lscratch6 | 90 | 5.7 |
| lscratch7 | 90 | 5.7 |
| OCF Maximum Lustre Bandwidths (GB/s) | ||||
|---|---|---|---|---|
| OCF System (CZ) | lscratchd | lscratche | lscratchf | lscratchv |
| Ansel | 12 | 12 | - | - |
| Aztec | .125 | .125 | - | - |
| Cab | 48 | 48 | - | - |
| Catalyst | - | - | 36 | - |
| Edge | 10 | 10 | - | 10 |
| Herd | 1.25 | 1.25 | - | - |
| OSLIC | 3 | 3 | 3 | 3 |
| Sierra | 48 | 48 | - | - |
| Vulcan | - | - | - | 108 |
| OCF System (RZ) | lscratchrzb | |||
| RZCereal | 10 | |||
| RZMerl | 12 | |||
| RZGPU | 12 | |||
| RZSLIC | 3 | |||
| RZuSeq | 16 | |||
| RZZeus | 10 | |||
| SCF Maximum Lustre Bandwidths (GB/s) | ||||
|---|---|---|---|---|
| SCF System | lscratch1 | lscratch3 | lscratch6 | lscratch7 |
| CSLIC | 3 | 3 | 3 | 3 |
| Graph | - | - | 15 | 15 |
| Inca | - | - | .125 | .125 |
| Juno | - | - | 40 | 40 |
| Max | 48 | - | - | - |
| Muir | - | - | 40 | 40 |
| Sequioa | 850 | - | - | - |
| Zin | - | 90 | - | - |
Maximum bandwidth limits are determined by either file system disk size, network/Lustre router speed, or nodes (i.e., single node job scheduling).
Users are reminded of the following:
- When moving files between Lustre file systems, always log in to the SLIC system on that network (OSLIC, RZSLIC or CSLIC).
- MyLC has a portlet called “scratch file system usage” that shows current Lustre file system usage. Just type mylc in the address bar of any web browser to access MyLC.
- Lustre file systems are not backed up. Users should copy important files to archival storage. Files older than 60 days are subject to purge without notice.
- Subverting the purge policy by touching all of your files will result in a loss of compute privileges.
- Performance degrades as a file system fills up. Deleting or archiving files not in use will keep the file system performing well.