Datahub & DSMLP: Overview¶
UC San Diego's Datahub and Data Science & Machine Learning Platform (DSMLP) provide thousands of undergraduates, graduate students, and their instructors with access to Jupyter, RStudio, and other advanced computational resources for scheduled classes, formal independent study, and student projects. The same cluster carries research work, where it is also referred to as the Research Cluster.
The platforms are used in courses ranging from introductory Data Science lectures, to graduate-level applied machine-learning, to curricula in Biology, Music, Social Sciences, and Public Health. Over AY 2025-26, more than 140 classes, 75 instructors, and 18,000 student enrollments were hosted.
Datahub is the web interface, at datahub.ucsd.edu. DSMLP is the cluster beneath it, and is also reachable from a terminal. They are not separate services, and the same files are present in both.
Please reach out to our team with any questions or feedback.
- Email: datahub@ucsd.edu
- 1:1 Consultation (instructors, TAs, and Technical Points of Contact): https://ucsd-datahub.youcanbook.me/
Changing in Fall 2026: GPU Reservations¶
From Fall 2026, access to GPUs is managed by a reservation system. A web interface books a specific window — "Tuesday 9am-7pm, 4× X-Large" — and for the length of that window the capacity is held and sessions are admitted ahead of the walk-up queue. Ad-hoc, on-demand use continues to be supported.
On-demand launches draw on the Service Unit budget exactly as a booked window would. Launching an eligible session is what authorizes that spend. Anyone using GPUs at all should read On-Demand Leases Charge Budget before the first launch of the term.
Where to Begin¶
| Audience | Please see | Which assumes |
|---|---|---|
| A student enrolled in a course that uses Datahub | Using Datahub in a Course | A web browser and nothing else |
That same student, where the course requires SSH, launch.sh, batch jobs, VS Code, or a GPU class the course menu does not offer |
Working from the Command Line | A terminal and ssh |
| Teaching or assisting with a course | Teaching with Datahub & DSMLP | A browser for most tasks |
| Undertaking a personal project, independent study, capstone, or club or team project | Projects & Independent Study | Familiarity with a shell |
| A graduate student, postdoc, undergraduate, or staff researcher working without a lab workspace | Research on DSMLP | Familiarity with a shell |
| Faculty arranging access for a research group | Setting Up a Research Lab | Some steps are performed by our staff |
Coursework and a personal project are two separate workspaces, with different eligibility, different resources, and different access lifetimes. Please read the course pages for coursework and the project page for the project.
Workspaces¶
Users are divided into groups called workspaces: one per course, lab, or general catch-all population. Workspaces anchor all cluster configuration: rosters, storage, container images, GPU access, and storage and GPU quotas.
A member may belong to several workspaces at once — a TA for one course, a student in another, a member of a lab — and picks the context to work in at launch time. A member of more than one workspace has a home directory in each. → Workspaces & Storage
Standard Features¶
Job environment. All jobs, whether web/Jupyter, CLI, or batch, present a compute environment derived from the member's account and workspace context together with any job-specific configuration. Course sessions typically begin at 2 CPU cores and 4GB RAM; upper limits are based on class size, demand, and capacity.
GPUs, organized into five classes by size rather than by hardware model, from approximately 6GB to approximately 96GB. Each workspace is given access to one or more classes, matching anticipated work. → GPU Classes
Storage. A per-user, per-workspace home directory; a shared area readable by everyone in the workspace; and a per-user, cluster-wide private area available in every workspace. Course home directories are typically 5-10GB; research accounts are substantially larger. Optional mounts of external storage are available. → Directories, Quotas & Cleaning Up
Standard Software Images — curated environments covering the widely-used Python, R, and Julia data analysis stacks, a CUDA/GPU-enabled image with TensorFlow and PyTorch, and an RStudio image. Courses and labs may further customize or pin their own. → Standard Images
Three ways to work: in a browser, at a terminal over SSH, or with a desktop editor such as VS Code connected to a cluster container. → Access
Three Things Worth Knowing Early¶
Containers run unprivileged, under the member's own UID, with no root or sudo
inside the container. sudo apt-get install will not work; this is by design and
is not a fault to report. → The Hard Boundary
The login node is not for computing. It exists to launch jobs and move files. → The Login Node
An idle GPU session still holds its GPU, and is reclaimed once it has been idle long enough. → Idle Culling
Support¶
| Topic | Please contact |
|---|---|
| A course a student is enrolled in | The instructor or TA, first |
| Datahub or DSMLP itself | datahub@ucsd.edu, or the IT Service Desk |
| The Research Cluster, or Universal Scale Storage | rcd-support@ucsd.edu |
| Which platform is appropriate for a research project | Research IT |
We aim to resolve individual user issues within 1-2 business days. Urgent or broadly-scoped problems may be escalated through the IT Service Desk; a ticket should say plainly what is affected and how many people. → Getting Help
Caveats & Limitations¶
No Sensitive Data: Datahub is not engineered to protect highly-sensitive data such as clinical records or export-controlled information ("P4" per University of California classification levels) and must not be used for such purposes. Legally- or contractually-protected information ("P3") may be permitted after review; note that depending on the nature of the data, vetting may take 4-6 weeks or longer. → Policy
Shared Compute Resources: Datahub and DSMLP system resources are shared among all courses and research groups. Demand for resources, in particular for GPUs, may exceed capacity at peak hours during 10th and Finals Weeks or at assignment deadlines. From Fall 2026, reservations, per-group quotas and Service Unit budgets are the mechanisms by which that contention is managed. → GPU Access
Scheduled Maintenance: Datahub may be unavailable Tuesdays, 6-8AM for installation of time-sensitive updates or security patches. In practice this work is usually limited to a subset of worker nodes, in which case running jobs are unaffected. Infrequent 'Critical' updates may require downtime outside of this timeframe.
Availability and Reliability: Datahub and DSMLP were designed with student workloads in mind, deliberately trading some of the costly redundancy typical of financial or health settings for additional capacity and capability. As such, they should not be used to host externally-available services or applications except as required for coursework or projects. (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.)
Appropriate Use: The campus-wide IT Acceptable Use Policy applies to use of Datahub and DSMLP, including prohibitions on commercial or political activity, hacking or cyberstalking, and other types of unwelcome behavior.
If you still have questions or need additional assistance, email us at datahub@ucsd.edu or submit a ticket to the ITS Service Desk.