Teaching with Datahub and DSMLP¶

This page is for instructors and TAs teaching a course on Datahub and DSMLP. It covers the course request timeline, standard features and customization, support and technical consultation, and GPU access for a course. Most tasks require only a browser, and course-specific customization requires a shell.

Course Timeline¶

Requesting a Course¶

Instructors, TAs, and departmental staff request Datahub and DSMLP access for a course through the Specialized Instructional Computing Course Request form. Submission opens in the 2nd week of the previous term. For Summer and Fall classes, submission opens in the 2nd week of Spring.

Requests must be submitted 4 weeks before the start of instruction, including the designation of a Technical Point of Contact (TPOC) where one is required, as described in Technical Point of Contact. Late requests receive lower priority and are reviewed as time permits. Setup and availability to students depend on the complexity of the course and the overall request load, and for a late request students may not gain access until as late as 4th week.

Course Setup and Instructor Access¶

Course setup and instructor access open 4-5 weeks before the start of instruction. Earlier setup is available on request. During this period the instructor, TAs, and other course staff test features, validate assignments, and make customizations. Availability of 1:1 Consultation for customization is limited in the final weeks of each term, as described in 1:1 Consultation.

Student Access¶

Student access follows TSS course rosters. TSS is the system formerly called TritonLink. Roster loading and the timing of add and drop changes are given in Students Enrolled in a Course. Auditors and observers are not on the roster, and their access is managed through Canvas. Concurrent Enrollment students receive access through Extended Studies staff.

Access After the Term¶

Instructors and students retain access for one additional quarter beyond the instructional term, ignoring Summer for Spring courses, as described in One Additional Quarter. Instructors may request that individual accounts remain active longer for Incomplete grade resolution, academic integrity proceedings, course development or hand-off, or similar circumstances, as described in Extending Access for an Individual.

Course environments are purged one quarter after account deactivation, that is, two quarters after the course, excluding Summer. Two things are excluded from the purge: individual accounts extended on request, and instructor and course-wide files archived on request. Archiving, reviving a previously archived class environment, and making an archive available for download are requests to ITS, described in Archiving on Request.

Independent Study and Research¶

For-credit independent study, capstones, and similar student projects request ongoing access through the Independent Study Request form. Direct those students to Projects & Independent Study. Research IT connects faculty, staff, and student researchers with compute platforms for non-credit research.

Standard Features & Standard Software Images¶

Datahub offers a set of curated software environments that are sufficient for many courses and use cases. These environments, with the session resources, course file distribution, grading tools, and individual customization support that accompany them, are the standard features. Questions, configuration assistance, malfunctions, and errors relating to standard features receive priority support from IT Services through the routes in Support & Technical Consultation.

Standard Software Images¶

The standard software images are accessible through web-based Jupyter notebooks and from the command line. They are described in Standard Images. The rstudio-notebook image derives from datascience-notebook and is not GPU-enabled.

Session Resources¶

CPU, memory, and NVIDIA GPU allocations are flexible, beginning at 2 CPU cores and 4GB RAM dedicated to each student session, as described in The Browser Session. Upper limits are based on class size, demand, and capacity.

Course Files and Storage¶

Course-specific file and dataset distribution is available, with the size limits given in Asking for a Dataset to Be Staged. Per-student storage allocations are given in Workspace and Personal Quotas, and course storage generally in Workspaces & Storage. Inform ITS where assignments generate significant per-student output.

Grading Tools¶

Tools for assignment distribution, collection, and grading, including roster integration with Canvas, are described in Grading.

Individual Customizations¶

Minor per-individual customizations within the standard software images receive limited support, as described in Customizing an Environment.

Course-Specific Customization¶

Instructors may supplement the standard software images by adding language modules (for example, Python or R libraries), adding system-level packages (compilers, utilities, and similar), or making more extensive modifications to meet the needs of a course. The procedure is described in Building & Publishing a Custom Image.

Highly complex use cases, such as those incorporating clustered services such as Spark or Postgres, or software not derived from the standard images, are regularly used within Datahub and DSMLP. They require substantially more time and expertise from the instructor or TPOC than ordinary customizations, as described in Complex Customizations & Experimental Features.

Technical Point of Contact¶

The instructor, a designated Technical Point of Contact (TPOC), or both must take the lead in the installation, configuration, and student use of course-specific features. ITS technical staff support customization through 1:1 Consultation.

The instructor or TPOC is expected to have basic familiarity with Unix commands such as ssh, mkdir, and chmod, and to be proficient with the intended core platform (for example, Python or R) in a desktop Mac or PC setting. ITS provides guidance and basic training on the Datahub and DSMLP environment and on system-specific procedures.

Pinning an Image¶

A course may pin a specific image so that students are not moved by a quarterly update partway through a project, as described in Pinning a Workspace.

Support & Technical Consultation¶

IT Services provides prioritized support for standard features and functionality through the Datahub & DSMLP Documentation, the IT Service Desk, and 1:1 Consultation. Questions and feedback about Datahub and DSMLP go to datahub@ucsd.edu.

IT Service Desk¶

The IT Service Desk is reached by phone, web, or email for the following.

Effort is prioritized by the number of courses and students affected and the overall impact on instruction. Note in the ticket where an issue affects a whole class or falls on an exam. Students route functionality concerns through their instructor or TA. In-depth questions go to 1:1 Consultation.

Response targets, including the incident tier for instructors, are given in Response Targets. Urgent problems arising outside business hours may be escalated through the Service Desk.

1:1 Consultation¶

1:1 Consultation appointments connect instructors, TAs, and TPOCs with ITS technical experts for real-time guidance or assistance on standard features, course-specific customizations, and complex or experimental features. Topics include:

• In-depth questions about the configuration or usage of standard and core features
• Discussion not well suited to ticket-based interaction
• Functional or technical issues outside the standard and core features
• First-time use of a custom container derived from a standard software image
• Customization options and the selection of additional packages and tools
• Functional or technical issues with course-specific features, such as build errors and version conflicts
• Questions and training on the customization process, such as the use of git, GitHub, tags, and Actions, or the installation of language and system packages
• Maintenance following quarterly updates to the standard software images

At Spring 2026 staffing levels, each course may request up to 6 hours of 1:1 Consultation per term. Availability is reduced in the final weeks of each term.

Complex Customizations & Experimental Features¶

The compute clusters underlying Datahub and DSMLP can host complex or novel customizations that fall outside the normal bounds of ITS support. A customization falls outside those bounds when, for example, installing or integrating it may exercise untested or seldom-used features of Kubernetes, Docker, or Linux, or student use of it may require sophisticated technical expertise or extensive assistance.

Available capabilities considered complex or experimental include:

• Instructor or TA containers not derived from a standard software image
• Student-built customized containers, of any derivation
• Background batch processing and analysis pipelines
• MATLAB (Jupyter kernels or web UI), GNU Octave, or similar complex applications
• Spark clusters
• ArcGIS integration
• Postgres and other persistent services

Incorporating these or similar features into coursework requires the instructor or TPOC to invest significant time before and during instruction, first to become independently familiar with the underlying technologies and then to serve as primary support for students' use of them.

ITS provides technical guidance for these features but, without advance agreement, is not responsible for implementation or usage. Schedule a 1:1 Consultation at least one full quarter before the planned use to discuss feasibility.

Visual Studio Code¶

Visual Studio Code is not among the complex or experimental features. It is supported and widely used. The supported configuration is Remote-SSH over a ProxyCommand, as described in Remote Editor Setup.

GPU Access for a Course¶

From Fall 2026, GPU access is managed by a reservation system, described in GPU Access. Its web app is at reserve.dsmlp.ucsd.edu.

Quotas and Service Unit Budgets¶

A course holds a GPU quota, the number of GPUs of each class it may hold at once. Its students hold Service Unit (SU) budgets, which divide that capacity across the roster. A course workspace's defaults, including its weekly budget, its booking horizon, and its length cap, are changed at the instructor's request by ticket. Quotas are described in Quotas, Cohorts & Availability and budgets in Service Units & Budgets.

Deadlines and the Quarterly Survey¶

Quotas are date-aware, so a course's share may be raised for the span of a project deadline and revert afterward, as described in Date-Based Quota Changes. The quarterly survey asks instructors and TAs about assignment scope, GPU sizes, and deadlines. The responses allow a course's quota to be raised in advance, for example for week 9. Report assignment deadlines in the survey.

Course Calendar¶

As workspace managers, instructors and TAs see every reservation in the course on the reservation app's Group Reservations page. They may book on a student's behalf, cancel a student's booking, and waive penalties, as described in Managing a Group. In a course whose roster is loaded automatically, the instructor, TAs, and graders are all workspace managers.

Assisting a Student at the Service Unit Limit¶

Request a budget change by ticket to datahub@ucsd.edu. The change itself is an administrative action.

Budgets are described in Service Units & Budgets, and the request in Administrative Requests.

Waiving a Cancellation Charge¶

A student who misses a booked window is assessed a cancellation charge. Instructors and TAs may waive the charge where warranted, on the Group Reservations page, when cancelling or afterwards. See Having a Charge Waived.

Manager Reports¶

Manager reports cover reservations by group, peak simultaneous use by class, reserved hours, and effective limits, as described in What the Reports Cover. Three of the four reports display cluster-wide data rather than data for the course alone.

Reservation Events¶

The reservation system reports on each student's GPU session with Kubernetes events. Datahub shows them while a session starts, and kubectl describe pod shows them from the command line. A student's question often quotes one. Each is defined in Reservation Events:

• While a session waits: WaitingForReservation, ReservationFull, ReservationTooSmall, OnDemandLeaseDenied, OnDemandLeaseRejected, OnDemandAdmissionPaused, UnknownGpuClass, NoReservation, AnnotationIgnored.
• When it is admitted: RuntimeGuaranteed, OverstayRelinked, BestEffortAdmitted.
• When it is stopped: Preempted, ReservationCancelled, ReservationReassigned.

Advising Students¶

Two conditions account for most of the student questions that reach course staff. Launching a GPU session draws on the student's budget whether or not a calendar is opened, as described in On-Demand Lease Charges, and a GPU session that stops using its GPU is reclaimed, as described in What Counts as Idle. Both are covered for students in Using Datahub in a Course, for browser users, and in Working from the Command Line, for work that requires a terminal.

Maintenance and Policy¶

Under the University of California classification levels, highly sensitive P4 data, such as clinical records or export-controlled information, is prohibited on Datahub and DSMLP, and legally or contractually protected P3 data may be permitted after review, as described in Data Classification.

Datahub may be unavailable during scheduled maintenance for time-sensitive updates or security patches. The schedule is given in Scheduled Maintenance.

Other conditions of use, covering shared compute resources, availability and reliability, and appropriate use, are on Policy, which also covers Self-Supporting Programs.

Report a problem with this page