We asked the MsPASS team to share what has changed since the course first came to GeoLab in 2024 and what those changes have made possible for teaching and seismic data workflows.
In 2024, Massively Parallel Analysis System for Seismologists (MsPASS) became the first technical course hosted in the GeoLab environment. At the time, GeoLab was still in its pilot phase, and we were testing whether the platform could support computationally demanding seismic data workflows in the cloud.
This July, Using MsPASS for Data Processing on HPC and Cloud Systems returned to GeoLab for its third year. Since that first course, there have been major changes to GeoLab, MsPASS, and the broader IT environment of the community that led to changes in class content and educational focus. GeoLab is now open to the community, the NSF NGF seismic archive is moving into cloud storage, and we’ve had two years to improve both the computing environment and how we prepare students to use it.
The movement of the seismic waveform archive to the cloud made it possible to develop data exercises that directly utilize the waveform archives. Previously, we had to use preassembled tutorial datasets. This year, participants ran a Jupyter notebook that extracted an event-based dataset directly from the continuous archives on Amazon S3. The notebooks used for that process are also a starting point for students and others in the community to access waveform data from GeoLab and are available through the MsPASS tutorial repository on GitHub.
The course also demonstrated the performance advantage of accessing waveform data directly from cloud storage. In the 2026 exercises, direct access to the S3 waveform archive was roughly three orders of magnitude faster than comparable single-threaded web services requests. The workflow also provides a more robust way to handle large data requests by tracking what is successfully retrieved rather than risking unnoticed data loss during a large download. With additional computing resources beyond GeoLab’s current four-CPU limit, even larger performance gains may be possible.
Beyond performance, an important result of these changes is their potential to enable more reproducible scientific results. Assembling and processing waveform data consistently can be difficult. We still have a way to go, but the ability to create a workflow that runs on GeoLab to assemble a dataset for a project is one important element of addressing that problem for the future.
MsPASS itself has continued to evolve. Designed for large-scale seismic data processing, the framework combines waveform and metadata management with parallel workflows that can scale across computing environments. Two technical developments shaped the 2026 course. The first improved how MsPASS works with client-server systems such as Amazon S3 and MongoDB in parallel-processing workflows. The second addressed memory-management issues that had made large datasets difficult to process with Dask. The MsPASS team developed a new workflow manager to handle those datasets more effectively, and students in the 2026 course were introduced to this new approach in GeoLab.
The emergence of increasingly powerful AI-based tools has also influenced the educational focus of the course. Using these tools helped us focus more on teaching fundamentals and reduce the tendency to bog down in the “how-to” details when teaching a complex software package like MsPASS. At the same time, using AI to fill in those details depends on the availability of extensive, open-source documentation. MsPASS supports that use through an extensive User Manual and tutorial GitHub repository.
The 2024 course gave us a blueprint for supporting technical courses in GeoLab. Since then, we’ve added structured pre-course preparation and workflows that connect students directly to cloud-hosted data and computing resources. MsPASS was the first course to help us test that model, and it continues to help us improve it.