By Hurray, hurray, another conference day! This time I visited the DevOpsDays Zurich in Winterthur, Switzerland.
The DevOpsDays Zurich conference already exists for many years. But it took me until this year that I've finally made it. Quite surprising actually, as it's basically just in front of my door (a short train ride away).
Note: Although the conference is called DevOpsDays Zurich, the event takes place in Winterthur. Winterthur is a city in the canton (state) of Zurich in Switzerland.
Here is a quick review of this year's DevOpsDays Zurich, including some of the talks and presentations.
The Opening Ceremony
The conference kicked off with registering at the front desk and coffee and Gipfeli (croissants). Due to the virtual tickets, which could be saved in a digital wallet on a smartphone, it just took a second to scan the tickets. Registering: done!
Head over to the queue of the coffee machine. Because everyone thought there's only one coffee machine. But it turned out there were a couple of machines around – just not at the entrance. Lesson learned, for next time.
Dirk Lehmann took the role as master of ceremonies – and it seemed he was born to do that. He explained the code of conduct and basically reminded everyone: "We're nice people. Let's be nice to each other.".
Dirk also went through the agenda, explained the concept behind the "Open Spaces" gatherings and finally announced the keynote.
The Keynote: A real AI use case
And with this it was time for the keynote. Dr. Marc Schuh from TNG Consulting presented a very interesting case: Using Artificial Intelligence in the real world.
The challenge in question was the following: Professional kayak athletes are filmed by their coaches and the position of the paddle compared to the waterline is calculated. There are best practices and "text book" angles. The closer the angle, used by the athlete, to the text book guidelines, the faster the athlete advances with the kayak.
To figure this out, the coach needs to go through a recorded video frame by frame and mark/label the frames where the paddle changes position (for example submerging under water). This is a tedious task, as one can imagine. Could this be improved with AI?
Marc answered this question in a perfectly worded answer: It depends. (this became a running-gag being re-used in the following talks). Of course AI could be used for it. But AI needs to data – a lot of data. And as it turned out, no AI LLM has even been trained to recognize kayak paddle, let alone the positions of the paddle.
After tediously training their AI model with 35'000 frames of paddle positions, Marc and his team finally could advance in their task. However they were hit with many obstacles: How do we find the paddle? In which direction is the kayak going? These were eventually all solved with very interesting approaches.
The big AHA! moment was when Marc revealed the success rate once the team (believed they) finished the work. A rather depressive 37% success rate! (The audience estimated a success rate between 50-80%). Marc revealed points they missed and lessons they had to learn to adjust their previous configurations.
After further improvements and additional data feeding from additional videos Marc's team eventually achieved a final success rate of 95%, which was in line with the manual work done by the athlete's coach. However the time saved by the sports coach was enormous – and therefore a huge success!
A very interesting talk, loved by the whole audience!
Going Serverless
The next talk was about "Serverless". Lena Fuhrimann started with a quick evolution from physical servers to virtual machines over to containers and one abstraction layer less: Serverless. Yet we all still know, there are still servers involved – but the responsibility of managing these is "handed" to the cloud provider.
In her talk she walked the audience through her experiences creating a serverless application, using AWS Lambda. The example was quite good and easy to understand. She also showed a really good (typical) example of running into an IAM permission problem.
Another very good hint was how to reduce the latency of a function by simply moving certain parts of the code outside the function.
Interesting were also the questions after the presentation. One of the question was if serverless functions could easily be backported again to an on-premise installation if needed. This question perfectly represented the currently growing skeptical thoughts towards the big (US) Cloud providers. According to Lena this is possible without too much effort, but depends on the type of application.
Internal Platforms to enhance the developer experience
The next talk was about building internal platforms to enhance the developer experience and was held by Christina Kraus from meshcloud in Germany.
Christina talked about experiences at large customers, with developers frustrated or de-movitated by either removing access or adding responsibilities.
That's why a clear and written down list of responsibilities in advance is helping to reduce frustration across all involved teams. A RA(S)CI matrix for example clearly shows who's in charge of what and also where to expect some overlaying and shared responsibilities.
I hope her advice will be picked up by team leaders and IT managers – as (I personally have seen in many companies) exactly these points left unmanaged, undocumented and undecided lead to frustration among different teams!
Improving pipelines
Back into a more technical topic was the talk from Daniel Raniz Raneland from factor10.
As most of the audience knew, CI/CD pipelines can take a while to run. Daniel pointed out some important do's and dont's for pipelines and how to improve their runtime speed.
A good example was to run lint after compilation of code – as compilation is always necessary to run tests later on. The linting just uses additional time, delaying the compilation. And therefore making the developer wait for the pipeline even longer.
Sometimes you don't need to run a whole pipeline including all tests for the whole application if only one part of the application (e.g. frontend) was changed. This can be done using filters in the pipeline jobs and was a very good advice.
But also other improvements were shown, including the caching of images used by the pipeline. Or splitting up the pipeline into several pipelines running in parallel. This often leads to the same amount of time in total – however the needed results to the developer are often shown several minutes earlier.
Building a global observability Dashboard at Google
In another talk, Robert Lehmann from Google showed how his team created a global observability dashboard for Google's internal services.
As one can imagine, and as Robert confirmed, there are hundreds of thousands of internal services running at Google. To allow the relevant teams insights into performance and metrics, an internal dashboard was built.
Robert talked about the challenges of fine-tuning these dashboards – primarily using variables and finding dependent services.
Don't burn out – get motivated!
An interesting non-technical talk about burnouts and how to fight them (and get back on track) was held by Sandra Czopek. It was one of the Ignigte talks, lasting a maximum of 5 minutes.
In her examples she referred to Walter White, the fictional character of breaking bad, reacting to a burnout or demotivated situation.
She gave good advice how to tackle and prevent burnouts, including taking real breaks and real vacations.
Hall discussions, location and evening event
One of the most underrated highlights are hall meetings and discussions. Where conference attendees just casually talk to each other during a break.
These are – for me anyway – the real benefits of such conferences, where you connect with fellow IT folks, meet new and old friends and former colleagues. After all, the IT world in Switzerland is "small" and you eventually cross paths here and there.
The location at the Alte Kaserne Winterthur is meanwhile almost too small for the conference. Luckily there are several floors which can be used and have a little bit more space around yourself. I found myself enjoying a few games of Street Fighter on an arcade with a fellow DevOps engineer, which was quite fun.
At the evening event in the Carribean Pearl location, old and new connections were revived and formed. This allowed to have discussions going into technical details or just having a casual talk about what's going on in life – while enjoying some offered drinks.
Big thanks to the organizers of DevOpsDays Zurich for the great organization and smooth event!
