TP Stack¶
Key Directories¶
Your admin will tell you the exact project directory name (e.g. hpc-lab-b4d7ff24).
| Path | Purpose | Backed up? |
|---|---|---|
~/ |
Your personal home — configs, scripts, small files | Yes |
/projects/YOUR_PROJECT/ |
Shared project storage — code, venvs, datasets | Yes |
/scratch/YOUR_PROJECT/ |
Temporary large files — intermediate results | No |
Warning
/scratch is NOT backed up and may be cleaned periodically. Download important results to your own machine.
Monitoring¶
View cluster health and GPU status:
URL: https://grafana-monitoring.tp-stack.co.uk
Log in with the credentials your admin provides.
| Dashboard | Shows |
|---|---|
| Node Exporter Full | CPU, RAM, disk, network per host |
| NVIDIA DCGM | GPU temperature, utilisation, memory |
Useful to check before submitting GPU jobs — is the GPU already in use? How much RAM is free?
Rules of the Road¶
-
Don't run computation on the login node. The login node is for editing, submitting, and checking results. Heavy work goes through
sbatchorsrun. -
Clean up
/scratch. It's shared and not backed up. Delete files you no longer need. -
Be a good neighbour. Only request the CPUs, memory, and GPU time your job actually needs.
-
Don't share your SSH key. Your private key (
~/.ssh/tpstack) is yours alone. Never email it or put it on a shared drive. -
Report problems. Something broken? Tell the admin. Don't try to fix infrastructure yourself.
-
Back up important results. Download final results to your own machine. The cluster is not permanent archival storage.