Resources & Limits¶
Disk Quota¶
The following table shows the disk quota allocation.
| Path | Size | Purpose |
|---|---|---|
/home/<userid> |
100GB | User home directory |
/scratch |
150GB | For storing temporary data. Please create a directory named with your COMP ID (i.e. execute command mkdir /scratch/<userid> and then chmod 700 /scratch/<userid to disallow others to access your files in scratch directory. Replace <userid> with your COMP ID) and store your files under /scratch/<userid> directory. Files outside /scratch/<userid> may be removed without notification.Do not store any permanent data. It will keep cleanup regularly Files or directories older than 60 days will be removed automatically. Please touch the files and directories regularly if you don't want your files or directories to be removed. The command to update access time of an directory recursively is find <path> -execdir touch -h -a '{}' +. (Replace <path> with the absolute path of the directory) |
SLURM QoS¶
In order to let more students use the HPC server concurrently, the below QoS rules are defined for different group of users. Please do not schedule a job that needs more resources than allowed.
| Resources | Postgraduate Students | Capstone Project Students | Research Students/Personnel* | Academic Staff |
|---|---|---|---|---|
| CPU Cores | 8 | 8 | 8 | 8 |
| Total GPUs | 2 | 2 | 2 | 2 |
| 1g.20gb | 2 | 2 | 2 | 2 |
| 2g.20gb | 2 | 2 | 2 | 2 |
| 3g.40gb | 1 | 1 | 1 | 1 |
| 4g.40gb | 1 | 1 | 1 | 1 |
| 7g.80gb | 1 | 1 | 1 | 1 |
| Running Jobs | 2 | 2 | 1 | 1 |
| Submitted Jobs | 4 | 4 | 3 | 3 |
| Max Job Duration (Days) | 3 | 3 | 3 | 3 |
* If research students/personnel of COMP would like to access the HPC servers, please apply it by filling out this form by their chief supervisor (PolyU domain login is required).
Note:
The above resources is reviewed regularly to fulfill the changes of requirements
1g.20gb, 2g.20gb, 3g.40gb, 4g.40gb, 7g.80gb are the GPU resources that you can refer in the SLURM commands. The first number indicates the number of GPU cores and the second number indicates the total RAM available for that GPU instance.
You can request more than one GPU resources in a job but please be reminded that only one GPU resource is visible to a process. You will need to handle it in your code if you want to use more than one GPU resources in your applications.