Day 1 of KubeCon2025 started with keynote sessions, followed by breakout sessions. This conference is run by Cloud Native Computing Foundation (CNCF), a project of the Linux Foundation, focused on Kubernetes and its various variants. Despite being mostly all in Georgia World Congress Center Building B, I still managed to get 15,000 steps.

The Long Line to get in

The point of getting my conference badge on Monday was to avoid the delays to get in on Tuesday for the Keynote sessions. Sadly, they put everyone in one long line that wrapped around like a ride at DisneyLand. People who already had badges in the same line as people who had to go through the registration process. It took nearly an hour to get in, and by then I had missed part of the Keynote sessions. 

On your conference badge, you can stick on a color sticker, Green that you are open to talk, and Red to be left alone, like you do at a Brazillian Churrascaria.  They also had stickers for your pronoun, but the writing was so small it was impossible to read when you were talking to someone.

Keynote Sessions

Fortunately, the Keynote sessions are scattered across Tuesday, Wednesday and Thursday mornings. The small portion I caught had two speakers, one wearing a Mexican Luchador mask. The key take-away? Kubernetes is difficult, and all of the audience are superheroes for the struggle supporting Kubernetes at your respective companies.

Taming the Complexity Beast: How Organizations are Rethinking Software Architecture and Deployment

This breakout sessions was a panel discussion. Kubernetes is all about decomposing "Monolithic applications" into smaller "micro services". But why does the IBM mainframe enjoy seven 9's availability, when Kubernetes deployments struggle to achieve four 9's? Kubenetes celebrated its 10-year anniversary, but people are now re-thinking the micro-services approach.

It is interesting that when people see my conference badge "IBM / Sponsor", they asked me about IBM mainframes, including LinuxONE, and why it is such as rock-solid platform for running Linux and Red Hat OpenShift.  I have spoken more about mainframes to people at this conference than the previous 10 months!

Free Books

There was a lot of swag given away at this conference, but I was surprised to see books! I got two books by Sam Bhagwat, co-founder and CEO of Mastra.ai, on AI Agents. The first titled "Principles of Building AI Agents", and the second "Patterns for Building AI Agents". Having just finished writing a containerized AI application in Python and JavaScript that used IBM watsonx API to process RFP documents for my group of sellers, I appreciate books like these.

Rook: Intro and Deep Dive with Ceph Storage

As an IBM speaker, I need to support my fellow colleagues. This session by IBM speakers Benamar Mekhissi and Blaine Gardner covered the underlying technologies of Rook and Ceph. IBM Storage Ceph is a commercial product based on the open source Ceph. Ceph provides a unified Block, Object and File (NAS) interface across a clustered set of nodes. Rook is a graduated CNCF project that brings Ceph to Kubernetes with an "Operator" and Custom Resource Definitions (CRD). IBM uses Storage Ceph for its IBM Fusion HCI device as a global data platform for Red Hat OpenShift, Kubernetes and AI workloads. Ceph can help with Disaster Recovery by replicating data to remote data centers.

Routing Stateful AI Workloads in Kubernetes

Kubernetes started out with stateless containers, riding the wave of "serverless" Function-as-a-Service (FaaS) like IBM Cloud Functions, Google App Engine or AWS Lambda. I have worked decades in the flash, disk and tape storage business, so I know reality is "stateful". The speakers Maroon Ayoub (IBM) and Michey Mehta (Red Hat) explained how Kubernetes falls short when it comes to Generative AI. The key metric is "KV Cache Hit Rate", where KV refers to Key/Value pairs, a popular NoSQL database format with only two columns where the first column represents a unique key, and the second column the value(s) associated with it. Failure to handle this could be a 10x cost factor for your AI-based containerized applications. The speakers presented "llm-d" as a way to help address this issue.

Poster sessions at the Poster Pavillion, Kube-Crawl, Expo Hall

I love poster sessions! When I ran IBM Technical University events, we would have these in the evening, and people would walk around, a white or red fruity beverage in hand, looking at each poster, stopping to talk to the speakers of the ones that interest them the most. The key is to capture your entire talk onto a single poster. I would help speakers use GIMP tool to make these. This was part of the "Kube-Crawl" at the Expo hall, providing an alternative to all the vendor booths hawking their wares. My only complaint was they ran out of red, and I had to switch to white for my second glass.

Outside it is a chilly cold in the 40s, but inside we are learning!  Sometimes, a conference is in a city with beautiful weather, and people are torn between attending sessions or enjoying the sunshine.  Not here in Atlanta, which is much colder than normal this week!

#Kubecon