GSAS Conference 2024

In this blog, we will share our experiences from attending the recent GSAS conference. Highlighting the sessions we found most interesting and overall impressions. At the end, we’ll wrap up with our thoughts on whether we found it worthwhile and if it’s something we’d consider for the future.

This year we decided to use our competence fund to try out a conference outside of Norway instead of going to NDC as usual (nothing against NDC though). We've had an increased interest in getting more in-depth knowledge about software architecture, and therefore decided on attending the Global Software Architecture Summit conference which for 2024 had a special focus on AI. The conference took place in beautiful Barcelona from 14th to 15th October and a day for workshops on the 16th, but we decided against attending the workshops.

Barcelona, with its' 5 million residents and amazing architecture, was interesting to roam around and explore in our downtime. Checking out the local cuisine and the famous Sagrada Familia.

Presentations took place in Axa Auditorium, which is quite centrally located, with over 600 people attending. You won't find many stands, "goodie bags", or people trying to reel you into a sales pitch, which for us was a positive. Some stands were f.ex selling books written by the presenters at the conference.

One of the event's highlights was the auction token system, which made networking fun and engaging. Attendees earned tokens by interacting with others and exchanging names, verified by the organizers and traded in for tokens, one for each unique name collected. This approach encouraged meaningful connections and made meeting new people, including approaching strangers, enjoyable and rewarding. At the end of each day, tokens were used to bid on auction items. This system not only incentivized socializing but also created a vibrant atmosphere, leaving a lasting impression.

The presentations were recorded but at the time of writing not all have been published, they can be found here. As there was only one track, it was possible to attend all the presentations. There were many interesting talks at the conference, below are summaries of the ones we found especially relevant.

What makes Software Architecture so inTRaCtAbLE?
Neal Ford

Neal had the first talk of the conference, going through more overarching subjects in software architecture, firstly the tradeoffs made when architecting a system and his First Law of Software Architecture: "Everything in software architecture is a trade-off". And the tradeoffs made when deciding whether to use a full event payload vs a key payload, a shared service vs a shared class library, when you should start breaking apart a service into multiple services, and the positives and negatives of using different service architectures.
Lastly he warned everyone to be ready for the big wave of bad code coming in with AI. His suggestion was to focus more on things like Architecture Tests to make sure that it does not affect the architecture when the wave washes in.

Ragatouille: A culinary adventure
Soham Dasgupta

Soham started off by introducing the movie Ratatouille and its similarities to Retrieval Augmented Generation (RAG) in how it follows the same strategies to deliver a good product. One of these steps was "Anton's approval" where Anton must approve the dish, similar to RAG evaluation where the answer given by the AI should be based on its context(relevance) while also be checked and scored according to Ground truth(accuracy). At the end of the presentation Soham showcased some examples on where to apply RAG, as seen in the image above.

Manufacturing Control with AI
Fernando Guevara

Fernando Guevara showcases how they implemented AI for their factories at PepsiCo and using the different steps to get there. They have multiple machines and sensors in their first layer that inputs massive amounts of data to their AI process. Their AI process then uses this data to give suggested instructions to the humans operating the machines to improve the manufacturing. The results of the instructions are then fed through to the AI again for it to analyze for the impact of the change.
This is a process they implemented at a factory making chips and it worked even better than expected, with improved production of 2%, which in manufacturing terms is a lot. He did bring up the caveat that when trying to implement it at a very similar factory making chips, the results were not good, as the process added spices at a different stage, and that was enough to throw the AI off.

Architecting for Sustainability
Eoin Woods

Eoin Woods started off by informing us why we should think of climate when creating software. ICT emissions are currently at 2-4% and by 2030 it may use 5%, so it's rising, and Microsoft increased their emission by 30% just last year, while for reference aviation is at 3%. Indicating that we as Software developers should think twice on emissions.
We were introduced to Green Software Foundation(GSF) which has some principles to follow, and showed some case studies where they managed to reduce emissions when using AI. As AI needs to run a lot of computation power, they could use a concept called Demand Shifting where they shift the workload to a different region according to metrics for green energy and save up to 40% emissions, for the price of a small delay. Another approach is to shift the workload to client devices, as almost anyone now has a pocket computer that's able to run big workloads which generally is on standby.
Eion finished off by showing us how we can start our journey to reduction of emissions for our software.

Complexity and Modularity
The Yin and Yang of Socie-Techincal Design
Vlad Khononov & Sonya Natanzon

Vlad Khononov unfortunately couldn't attend, so the presentation was held by Sonya Natanzon only.
Sonya went through the details of Conway's Law and what it means in practice. Conway's Law is "Organizations which design systems (in the broad sense used here) are constrained to produce designs which are copies of the communication structures of these organizations.".
She started off by introducing the concepts of a social system and its design. One of the topics was "What is complexity", which we found interesting as it touched on different practices for complexity(image) and Social/Organizational complexity, and the symptoms of both.
The organization needs to be organized according to what kind of system you want to create, so this is introduced by Vlad and Sonya as Modularity = Strength XOR Distance. So if you want High Cohesion, you need Low Distance and High Strength. If you want Loose Coupling, you need to have High Distance and Low Strength.

Wrapping up

The unique token system was refreshing and fun, adding an extra layer of excitement to our activities. Even though we didn't gather many tokens, the experience itself was enjoyable and memorable.

The food situation was decent overall. While the lunch wasn’t particularly substantial, there were many breaks with finger food servings. The variety of snacks kept us satisfied throughout the day, and the frequent breaks provided a nice opportunity to stretch our legs and chat with other attendees. Although a more substantial meal would have been nice, the setup worked well for the event’s format.

The high number of attendees led to inadequate air circulation in the auditorium, making the atmosphere stuffy and uncomfortable. Especially after each talk, opening the doors brought much-needed fresh air inside. This experience highlighted the need for better ventilation in future events.

We’re excited to explore larger conferences, so we probably won’t attend this seminar again. However, the organizers did a fantastic job selecting outstanding speakers, which greatly enhanced the overall experience, and we look forward to seeing them at future events.

Overall, the experience was positive. The organizers were friendly, and it was refreshing to attend a conference outside of Norway.

This concludes our trip to Barcelona and GSAS. Farewell from Sævar Thorisson & Henrik Bjørnland