Three separate obligations sit on everyone who uses Datahub and DSMLP: what the University permits on the platform, what campus policy permits on any IT resource, and what colleagues on a shared cluster are entitled to expect. This page covers all three.
P4 data is prohibited. Highly-sensitive information such as clinical records or export-controlled material must not be placed on Datahub or DSMLP. There is no review path and no exception.
P3 data may be permitted after review. Legally or contractually protected information is not categorically excluded. It requires a conversation before any of it reaches the cluster, and depending on the nature of the data, vetting may take 4-6 weeks or longer.
The University's classification levels are the authority on which level applies. The examples our own documentation gives are:
| Kind of data | Examples |
|---|---|
| Government classified or controlled | CUI, CTI, ITAR, FISMA |
| Health and personal information | PHI/HIPAA, IRB-controlled data, statutory PII |
| Student records | Other students' grades or academic records (FERPA) |
| Contractually protected | Information subject to certain Data Use Agreements |
Where to raise it. For instruction and coursework, email datahub@ucsd.edu. For research, Research IT at rcd-support@ucsd.edu. → Restricted & Licensed Datasets · Getting Help
The campus-wide IT Acceptable Use Policy applies here as it does to any campus IT resource, including its prohibitions on commercial or political activity, hacking or cyberstalking, and other types of unwelcome behavior. Nothing about this platform's academic purpose relaxes it.
Access is granted to an individual, not to that individual's collaborators.
An account, its home directory and any dataset granted with it are for the
account holder's own use and are not passed on. The workspace's public/ and
teams/ areas are the sanctioned route for sharing.
→ Inside the Workspace
Datahub and DSMLP should not be used to host externally-available services or applications, except as required for coursework or projects.
A compute node can be drained for patching, a pod can be preempted, and a session has a deadline. (This caveat applies primarily to the compute nodes executing user jobs; critical components such as networking, file storage and backups are maintained to Enterprise IT standards.)
A class demo, a project web application, a service an assignment requires: all in scope. A production service, a persistent public endpoint, or anything somebody outside the University would notice going down: not. → Port Forwarding & Reaching Services in a Container
Every one of these is a request rather than a rule.
Please stop sessions that are not in use. GPU cards are assigned to containers exclusively: a card attached to an idle session is unusable by anybody else. Stop the session with File → Hub Control Panel → Stop My Server, or delete the pod from the login node. Logging out, closing the tab and closing a laptop all leave it running. → Datahub in the Browser
Please do not treat idle culling as a cleanup service. The delay before it acts is measured in hours on a quiet cluster. → Idle Culling
Please debug on CPU before requesting a GPU. Both PyTorch and TensorFlow switch between CPU and GPU easily, and a CPU-only container is not the scarce thing. → Access
Please ask for the smallest thing that works — the smallest GPU class a model fits in, the memory a job needs rather than the maximum permitted, the hours it actually takes. A larger class is not faster for a model that already fits in a smaller one; it is only scarcer. → GPU Classes
Please do not compute on the login node. It is a jumpbox for launching jobs and moving files. Running a training script, a build or a long analysis there is prohibited. → The Login Node
Please do not keep a personal copy of a shared dataset. Read it where it lives. A per-student copy of a large corpus exhausts a home quota, and multiplied across a roster it exhausts rather more than that. → Shared Datasets
Please cancel a booking that will not be used. Cancelling in advance carries no penalty, and the capacity returns to the pool. → The Cancellation Penalty
Demand for GPUs may exceed capacity at peak hours during 10th and Finals Weeks or at assignment deadlines. Work shifted to a daytime or off-peak hour costs less. → Off-Peak Discounts
Self-supporting programs are welcome, and there is a cost attached. Self-supporting programs such as MAS and MBA may use Datahub and DSMLP for coursework or projects, and UC policy requires us to recover the associated direct and indirect costs. Please contact us at datahub@ucsd.edu to discuss it.
Note separately that storage above 1 TB is chargeable. Compute is not. → Mounting External Storage
On the instructional side, maintenance runs Tuesdays, 6-8 AM. In practice the work is usually limited to a subset of worker nodes, in which case running jobs are unaffected — but please do not plan a deadline into that window. Infrequent 'Critical' updates may require downtime outside it, in which case we notify instructors as soon as is practical.
The Research Cluster window is different in kind. It is quarterly, at the break between terms, comes with about 7 days' notice, and all running jobs are terminated. → Maintenance Closures
If you still have questions or need additional assistance, email us at datahub@ucsd.edu or submit a ticket to the ITS Service Desk.