1st Stuttgart Research Software Day
I arrived early in Stuttgart to join the first Stuttgart Research Software Day. Research Software Engineering seems to be quite big in Stuttgart. Not only is the HLRS, one of the three German tier 1 supercomputing centres, located there but also many large RSE projects seem to originate from Stuttgart. The University is well on their way of establishing a central RSE department. I had a great chat with the people from the bwRSE4HPC project who support researchers in Baden Württemberg to develop and deploy their software on bwHPC HPC systems. The other project worth investigating is the Unfold framework for EEG & Neuroimaging written in Julia. More posters can be found on the 1st Stuttgart Research Software Day Zenodo Community.
Day 1 deRSE26
The morning of the first day of the conference was slow and gave me a chance for a little wander through town. The first keynote was from Wolfgang Bangerth who spoke about what it takes to run a big successful open source mathematical library project. He is one of the main authors of the deal.II finite element library. The talk was fascinating. His position is that being a good programmer is a prerequisite. However, for large projects the people skills become extremely important. If communication between all stakeholders does not run smoothly the project will fail. The insight is perhaps not all that surprising but it was good to see it presented in such a nice way.
My talk on teaching digital literacy was last in the session. I was a little bit worried I had too little material. It worked out just fine. The talk was well received and sparked a number of discussions. The slides for the talk can be found on zenodo.
The poster session was in the evening. I was particularly interested in the BinAC2 - Cluster for Bioinformatics, Astrophysics and Geosciences poster. BinAC2 is part of bwHPC and supports among others bioinformatics and medical sciences. I chatted with my colleague about the DRIFT project which defines research facilitators. These facilitators help large research projects manage available resources and point them to relevant support such as data stewards and RSEs. The Helmholtz federated IT services presented their learning platform based on Open edX which looks really cool.
Day 2 deRSE26
Day 2 started with the keynote by Morane Gruenpeter from Software Heritage on digital sovereignty and forges. Software Heritage archives software projects from forges such as github or gitlab and assigns persistent identifiers. At the moment issues and merge requests are not archived but they are looking into extending their data model to include the discussions as well. AI generated code was obviously topical. They are already seeing an increase of the archive growth rate. They are looking into whether they can distinguish human from machine generated code. I suspect that will be very difficult.
After the keynote we had our fishbowl session on RSE pathways in Germany. The workshop room was well packed and we had some interesting discussions. We will attempt to produce a paper for the conference proceedings again. Toby Hodges from the Carpentries introduced our effort of coming up with AI programming courses. In the afternoon I did not have all that much choice of which talks to attend since I had to mark the presentations of some of the junior RSEs. The LSLAutoBIDS talk on automatically preparing neuroimaging data was particularly interesting.
In the evening we had the conference dinner in the Kursaal in Bad Cannstatt. The dinner was excellent and it got rather late with a stop in the pub on the way to the hotel.
Day 3 deRSE26
I almost made it to the first talk in the morning, the S Bahn thwarted me. Stuttgart public transport is somewhat in chaos at the moment with the transition to the new digital signalling system.
My pal Giacomo Peru from the University of Edinburgh gave a talk on the EVERSE reference framework for research software. This looks quite interesting and is something I should keep an eye on.
The final session I attended was on interactions with forges. Chung-hong Chan gave a super interesting talk on why it is so difficult for us to leave github. He took a sociologist's point of view and describes github as the field where we gather social and status capital. Giving up a community (or maybe just a bunch of followers) that we have cultivated is not easy. My perspective is that github is super convenient since everybody can have an account and CI/CD integration works out of the box without having to do anything complicated. I would very much like to see some form of federated forge where institutions can host their own forges which allow external collaboration via federation. Codeberg is working on federation.
The final talk was by Philipp Sommer on using templates for projects. The CD templates project looks really cool. The project supports a hierarchy of templates. Instances of templated projects get automatically updated when the templates change via merge request. I think this would work really well for my LaTeX presentation templates. Something to play with when I have a spare minute.
In the afternoon I got to visit the HLRS. We were split up into 4 groups. My group first got an overview of the HLRS. They have 27 permanent staff and 120 people work on projects. The current system, Hunter consumes about 4MW of electricity. It is a small system which the researchers can use to port their projects to the new system Herder for which a new building is being built. The systems get their performance from using AMD accelerators. Hence the need for the transition system. Herder will consume up to 8MW of electricity. Most of the energy is converted into heat. The warm water from the liquid cooling is heated up to 90 degrees C using heat pumps and piped into the local district heating network. The supercomputing system is essentially a heating plant. Unfortunately, it is less reliable as a source of heat energy than a traditional power plant. So there is the question of how to deal with backup heating for the district. The other thing that really impressed me was that they use three flywheels to smooth power fluctuations. In the case of a power cut they provide electricity for up to 30 seconds, enough time to take a checkpoint from which the simulations can be restarted once the power is back again. There is no point in having backup power since even a gas fired power plant could not spin up in time. The local topography is not suitable for a hydro pump storage plant.
After the presentation we got a tour of the machine room. First, we went past the first supercomputer, a Cray 2 from 1986. Today's smartwatches have more compute. At the time it was the biggest computer in Germany. The machine room was mostly empty. An air cooled NEC vector system was making an impressive amount of noise. Hunter, the current system is just a few racks long and makes no noise at all. Data are stored on a lustre system together with an 8PB tape library.
After the machine room tour we got to see the cave, a 3D visualisation facility. I hadn't been in one in ages. The system displayed a 3D model of the surrounding area. Various visualisations could be added. And we could fly/drive through the 3D world. The sudden descents produced somewhat wobbly knees. One visualisation used photogrammetric images gathered by plane. The model on the ground was not very good and looked more like a molten 3D nightmare scenario. The system can also be used to try out user interactions, for example move up a virtual ramp in a real wheelchair to operate the virtual ticket machine. They also offer augmented reality visualisation where a real 3D model of, e.g. a car, can be moved around on a table. A camera films the model and displays it on a screen together with whatever visualisation should be shown.
Conclusion
The deRSE26 conference in Stuttgart was great. I really enjoyed the chance to catch up with collaborators. Lots of interesting talks and posters provided food for thought. Some of the presentations and posters can be found on the deRSE26 Zenodo Community. Next year the conference will be held in Dortmund.