Web based notebooks for teaching, an experience at Universidad de Zaragoza
Miguel Angel Marco Buzunariz
Zaragoza, July 2020
How it started
- In 2011, decided to try SageMath for teaching
- It was hard to install, mostly on Windows
- But had a web based interface
- Got some funding for a server
- Started small, but eventually more instructors decided to use it
- Added a second server
Usual workflow
- The instructor of each course creates a worksheet and published it
- The students create a copy of the published sheet, work on it and share it with the instructor
- The instructor grades and and modifies the student’s worksheet
- Maybe add further work for the student to do
- The student does the extra corrections requested by the instructor
Problems that appeared
- Users had to be added manually
- The list of worksheets became too bloated
- Overloaded during peak times
- Specially during exams
- Suspect students trying to sabotage the exams
- Stopped development, newer versions dropped it.
New design
- We needed a new system with the following characteristics:
- Allow different users to log in, ideally using the university’s directory.
- Organize their worksheets in folders or similar.
- Persistent storage of the work of each user.
- Share worksheets with other users, at least in a similar fashion than the workflow we had in SageNB.
- Scalable
- Isolate the computations of each users, in such a way that one user cannot exhaust the available resources.
- The most evident answer was the Jupyter ecosystem
- JupyterHub with CAS authenticator
- JupyterLab includes a file browser with drag and drop support
- /home shared by NFS
- Started to work on a JupyterLab extension for that (far from ready yet, using email/moodle for the moment)
- Installed several machines behind a HAProxy (thanks to BiFi)
- Spawner runs user’s JupyterLab instance in a systemd container
New design
In the 2019/2020 academic year, we introduced this system (together with the old one)
Advantages
- More modern and maintained by a wide community
- More flexible:
- SageMath
- Python
- R
- Octave
- Maxima
- others (Julia, Coq)
Problems detected
- Barely enough manpower to develop and maintain it
- Could not cope with the peak demand of RAM (maybe Jupyter’s overhead?)
- Some students complained:
- In general, about the availability failures
- Engineering: “why do we have to learn yet another computing language just for math”
- Sciences (optincs/biotech): “This is too complicated”
- Mathematics/physics: better reception overall
- Again, scalibility problems during the exams
During COVID-19 lockdown
- It was a crazy time in general for everybody
- The network services in the University were stressed to the limit
- Most teaching methodologies and workflows had to be adapted
- But classes using these notebooks needed very little adaptation
Ideas for the future
- Try to get more hardware
- Move to a new design with just one Jupyterhub instance
- Develop a custom LXD spawner
Conclussions
- Jupyter based teaching has potential to be very useful
- They can be offered as a university-wide service to the students and teachers
- But that requires resources:
Personal thoughts
- General trend to move everything to “the cloud”
- More smartphones and tablets, less laptop/desktop computers
- Risk of losing the control of our data, and hence, our autonomy
- It is important to push our institutions to host their own services
- Based on free software
- And participate in its development
