I wanted to share my thoughts on the semi-annual confab that is Kubecon, this one the European edition. These are fairly candid takes, I can be critical or complementary, but given how important this space is to us, it is worthy of analysis.
Let’s get one thing out of the way. This was a superb Kubecon. The location was great, the weather was good, the venue was great – even the food was edible. It was, and felt, crowded (in a good way) and it was the busiest Friday we can remember. My only complaint would be that the CNCF seems a little too focused on the soft stuff and not enough on the business outcomes. The business outcomes are what fuel successful technology movements and while community is important (and we are part of that community), sometimes I worry that business outcomes get the way way back.
More importantly, however, the conversations are really maturing. We brought a deeply technical team to the show and we could have used one or two more during the busiest times. The delegates wanted to talk about details and we were happy to oblige. Here is a look at our visitor breakdown:
This maturity was reflected in the themes that we heard in our conversations – granted these are conversations with folks in MinIO’s booth so they tilt towards storage but we were so busy that we didn’t get out much.
Big Enterprise Deployments
As maturity grows, so does the ambition. We are hearing about deployments with 100s of clusters and PBs of data. We are hearing about edge cases that number in the 1000s of endpoints. We are hearing about highly distributed deployments and we are hearing about multi-cloud deployments. Our takeaways is that as the technology and knowledge base matures that enterprises are willing to take on more sophisticated architectures in service of the business goals. Yes, simplicity should always be a guiding philosophy but simple shouldn’t stand in the way of the outcome. Simple sophistication is achievable and we are starting to see more of that.
GUIs for the Win
Graphical user interfaces are overtaking CLI and that is driving the deployment wave for more and more enterprises. What this tells us, is that the concepts of containerization and orchestration have become deeply ingrained in the organization and that a broader audience is consuming the technology. Yes, you will always be able to do more from the CLI – but the goal should be to constantly narrow that gap. We are hyper-focused on just that goal, to narrow the gap for what you can do from our Operator interface. We have invested significantly over the past year to make the storage experience simple, powerful and secure. Given the interest in the Operator at the show we are going to write a companion piece on the latest additions.
Security Still Tops the List
Speaking of security – it remained a key topic at the show (this is a nice take). Securing your containers, securing your data (in-flight and at rest) were frequent topics of discussion. One of the areas where we have advanced the security agenda for Kubernetes is isolation. A lot of effort has gone into tenant isolation and requiring as few permissions as possible when deploying and managing pods. We want to keep the attack surface as small as possible. Another key aspect is encryption and key management. We have simplified the encryption setup to enable users to encrypt their data even when they haven’t set up their KMS infrastructure, yet. We believe users should be able to encrypt their data effortlessly from the beginning.
Both of these are immensely important to the overall goal of security. We have stated on numerous occasions – humans are the biggest security risk. Insider threats are one thing but just as much, if not more, security incidents are caused by configuration issues.
Keep security simple.
Performance at Scale + How that Differs from Performance and Scale
Performance matters and performance at scale matters the most. Performance is fairly workload independent. Yes, there are workloads in the data space (archival for example) that only put a mild premium on performance, but even those benefit from performance (small objects as an archival example, restore times as another). As the size of the Kubernetes workloads grows, so do the requirements to perform at scale. Performance on small data is fairly easy – in most cases the hardware does the work. Performance on PBs of data is different – this requires software that is optimized for these environments. The software should saturate the hardware – which is exactly what MinIO does at scale as demonstrated by the benchmarks we have published.
We welcome your thoughts on other themes you thought were impactful from the show and we look forward to seeing you in Chicago. We will bring even more t-shirts (we moved nearly 1K in Amsterdam) and some more game pads too.