Reference · RCC ClusterDocs
Safe everyday RCC practice
Security should support research rather than make ordinary work impossible. The following habits prevent most avoidable incidents without adding extra steps to every session.
Accounts
- Use your own human account.
- Add colleagues to the project instead of sharing credentials.
- Lab instruments use individually registered machine accounts restricted to their spool location; people must not use those accounts.
- Never register one SSH key, passkey or TOTP seed for several human accounts.
Credentials
- Keep SSH private keys on the device where they were generated.
- Prefer a FIDO-backed SSH key where a compatible authenticator is available.
- Keep SSH agent forwarding disabled and do not copy long-lived private keys between computers.
- Store passphrases in an approved password manager.
- Never paste credentials or private data into issue trackers, chat, Git, notebooks or support screenshots.
- Report a lost device or unexpected session promptly.
Endpoint trust
- A personally controlled endpoint is suitable for normal RCC SSH, VS Code and file-transfer workflows.
- A restricted institutionally managed endpoint must use the current RCC-approved connection method and may need support after local policy changes.
- An endpoint administered by an untrusted or actively hostile party is not suitable for sensitive RCC work, even when SSH encryption succeeds. Use an RCC-controlled workstation, virtual desktop, terminal or another trusted endpoint.
- Never approve an unexpected RCC server-identity warning. Stop and contact RCC support.
Cluster availability
- Use Slurm for computation.
- Avoid job storms, arrays created by mistake, tight retries and repeated connection loops.
- Do not recursively scan shared storage merely to count files.
- Check available capacity before writing large outputs.
- Stop and ask when a tool unexpectedly creates extreme load.
Web applications
- Do not implement a separate project password database.
- Do not expose a development server directly.
- Do not mount an entire project filesystem into a web application.
- Use the governed proxy, project authorization and curated data interfaces.
Biomedical data
- RCC may process biomedical research data when the approved project purpose and safeguards cover RCC; the data do not have to be anonymous.
- Removing names is not enough when genomic, imaging, free-text, date, rare-disease, or linkage information can still identify a person.
- Keep direct identifying fields and re-identification keys outside RCC.
- Complete Class 13; keep direct identifiers and re-identification keys outside RCC, and confirm that the project governance covers the biomedical data and intended use.