KubeCon Paris: edge, AI and la vie en cloud native

In case you missed it, last week was KubeCon Europe 2024, in Paris. Along with a dozen or more Spectronauts I was there, ready to take in the sights and sounds of my first KubeCon. And what a week it was. “La vie en cloud native” indeed!

Priyanka Sharma, the executive director of the CNCF, reported in her opening keynote that Paris was the biggest Kubecon ever with more than 12,000 people in attendance — and by the looks of the show floor, I'm sure that was true. 

The event was completely packed most of the time, and popular sessions had lines out the door. This year is the tenth birthday of Kubernetes, and I guess you could say the birthday party was well attended!

With more than 200 talks and other sessions happening across the week, no single blog can possibly sum up KubeCon. But from my perspective as a speaker, OSS contributor and attendee, a few key trends stood out: edge and AI most importantly.

Colo day? All about the edge 

Traditionally the first day of the show is dedicated to co-located events like ArgoCon, Istio Day, Cloud Native AI Day, and Kubernetes on Edge Day. 

Since I work on Kairos, an OS for running K8s at the edge, I naturally hung out at Edge Day.

Saad Malik, our CTO at Spectro Cloud, set the tone with his keynote showing how edge has gone from the lab to the real world, with examples from energy to agriculture — all being enabled by breakthroughs from the open-source community and the vibrant landscape of commercial vendors. 

Proving Saad’s point that edge is now out in the real world was a great talk fromYann Rotilio at the SNCF (French Railways), "Revolutionizing Railways: How Edge Technology Solves Decades-old IT Infrastructure Challenges". 

SNCF is using modern software in old trains to improve the experience of travelers and train companies alike — a great example of how we can stay sustainable and innovative at the same time (and ironically, I'm writing these words on a French train from Paris to Brussels). 

Personally, I was excited when Yann said they were inspired by Kairos to come up with their solution. This is, to me, one of the strengths of open-source: to be able to read the code and gather inspiration. 

When I wasn't attending a presentation, I was hanging out at the Spectro Cloud booth, hearing first hand the real experiences of the edge community. We were swamped! There’s so much interest in edge.

It turns out that the SNCF weren’t the only ones who had heard of the Kairos project: a team from Ericsson told me they consider Kairos to be the "best out-of-the-box solution on the market for Day 2 operations for Edge devices". This kind of market validation shows we’re on the right track!

The main event: AI for everyone

Wednesday was the start of the main event, and it was clear from the opening keynotes that AI was going to be a massive theme throughout.

Nvidia presented on making GPUs first class citizens in Kubernetes, Bloomberg talked about how it has replatformed its flagship AI workloads onto Kubernetes, there was a panel sketching out some patterns around ML/AI systems and processes… and much more. Frankly, across the agenda there may have been a bit too much AI, and it was hard to filter which sessions were meaningful and which were really AI clickbait.

"How to Choose the Best Kubernetes AI Edge Deployment Patterns for Your Use Case" definitely falls into the former category. Jacqueline Koehler and Myriam Fentanes Gutierrez explained the challenges of running AI on edge devices — but emphasized the need for it when it comes to critical use cases like cars, where hundreds of milliseconds can already be too long to make a call to a data center. 

Their solution adopts a GitOps approach, which has already been battle-tested. In my opinion, this kind of solution will allow us to see newer and better models out on the market and in all of our smart devices. One insight they shared is that most companies are finding it difficult to manage devices at the edge… a problem that we’re working hard to solve at Spectro Cloud with our Palette Edge platform.

One clear trend across the sessions and the corridor conversations was how to apply existing cloud native tech to AI use cases, for example storing models in OCI registries, using Knative and Kserve for inference and using Argo CD to deploy models. 

This was balanced by discussion of newer projects that are spinning up to address the unique challenges that AI brings, including Akri, a CNCF sandbox project that allows edge devices to be discovered and represented as Kubernetes resources in a cluster.

Perhaps the most encouraging thing to see is that as a community we are starting to build up mental models that we can use to communicate and innovate around. 

There was consistency in defining three main classes of AI/ML activities (model training, model refinement and model inference). We saw people talking about architectural and operation patterns, for example CI/CD, operationalizing the connection between model building and production deployments. Priyanka referred to Hugging Face as the “Github of ML”. I heard someone pose the question of "what is the LAMP stack for AI model delivery?" 

Through these discussions you can see that the community is starting to think about AI/ML through software engineering discipline, and this is vitally important.

The corridor track was buzzing

Towards the end of the show I spent some time on the general ‘corridor track’ and the booth space. It was absolutely buzzing, particularly at the Spectro stand where there was a heady mix of blue wigs and hardcore edge demos, and guest appearances from some of our partners, Natilik and DeARX.

We were lucky enough to get three talks accepted at KubeCon, and I urge you to check out the replays below.

Two node HA at the edge

First up, my colleagues Tyler and Pedro explored a new architecture for delivering HA at the edge with only two nodes. The demo gods didn’t quite play ball, but the room was full and there were lots of questions — clearly the community is keen to learn how to cut costs and drive sustainability in edge deployments, without sacrificing high availability!

The architecture in this talk is extremely interesting, stripping out the default etcd state store. Oh, and did I mention that Kairos is a big part of it too? Watch the replay right here.

Taking Kubernetes to thousands of dental offices

Next up, Bogdan Mitrea from Spectro customer Dentsply Sirona was on stage in another packed room, explaining how edge Kubernetes “Won't Hurt a Bit". 

Bogdan shared how Dentsply Sirona achieved a single node Kubernetes solution that is easy to deploy and manage through Day-2 operations, while remaining secure. This is no small achievement when you see their customer dental offices are all over the world. 

For us in the Kairos team, Dentsply Sirona’s use case is a great fit for our recently released Trusted Boot feature, which makes use of Secure Boot, Measured Boot, TPM and encryption to make devices at the edge secure. 

The session was really insightful and clearly many other companies out there are looking to do similar projects, because many people stayed to ask questions.

Watch the replay here: 

Special purpose OS panel

At the end of Friday afternoon, when the attendees were tired and ready to go home, we had one more session: a panel on special purpose operating systems. A last burst of energy was needed, and we brought it in spades!

Me and my colleagues from the Special Purpose Operating Systems Work Group (sp-os-wg) had an interactive session talking about what SPOSes are and answering many questions from those who attended.

I was in great company with experts from AWS, Microsoft, Google and Unikraft up on stage, with representatives from Talos and EveOS in the audience, and moderation courtesy of Thilo Fromm also from Microsoft, who did a fantastic job keeping the audience engaged. 

These kinds of initiatives are important because it helps users understand why these projects exist, why we have so many different options — and how that makes us a stronger community instead of rivals. 

Watch the panel here:

See you in London?

On behalf of the Kairos project, Spectro Cloud and all our partners and customers, I think it’s safe to say that Paris was the best KubeCon Europe yet, full of learnings and great conversations about AI and everything else. Next year the destination will be London. Will you join us there as we kick off Kubernetes’ second decade?