[KubeCon Day 4] Participation Report
Suzuki (@sZma5a)is.
#SRG(Service Reliability Group) mainly provides cross-sectional support for the infrastructure of our media services, and is responsible for improving existing services, launching new ones, and contributing to OSS.
This article is a report on our participation in KubeCon, introducing the contents of the keynote and sessions on Day 4.
Introduction
KubeCon + CloudNativeCon North America has finally reached its final day, Day 4.
On the final day, there were many important announcements that will shape the future of cloud-native technologies, as well as concrete sessions that solve real-world challenges.
In this article, I'll summarize the Day 4 keynote and report on the sessions that I personally found particularly interesting.
Keynote
The final day's keynote looked back on previous conferences and discussed a variety of themes looking ahead to the future of cloud-native technology.
Evolving Platform Engineering and Controlling Complexity
He introduced the challenges and success factors of platform engineering, pointing out that while template distribution is convenient, it also tends to distribute the operational burden.
He emphasized that a successful platform should adhere to three principles: "self-service via API," "continuous adaptation to the business," and "treated as a managed service," and that it should be built by identifying the unique value of your company while utilizing market services.
Kubernetes supports the AI era
Next, several companies spoke about the relationship between Kubernetes and AI, sharing case studies.
They discussed the importance of Kubernetes as a high-performance, secure, and open foundation for building large-scale AI platforms.
In particular, specific technology demos were shown, including AI inference in multi-cluster environments and GPU resource optimization, and it was explained how open source technologies and contributions from the CNCF community are supporting companies' rapid AI development.
Standardizing Kubernetes networking
In the networking session, it was introduced how Kubernetes' networking functions are evolving to support new workloads such as AI and HPC (high-performance computing).
One feature that attracted particular attention was Dynamic Resource Allocation (DRA).
This is a new API for allocating specialized hardware resources such as GPUs and high-performance NICs to Pods in a more flexible and standard way.
This is expected to unify resource management mechanisms, which have previously been implemented independently, and enable centralized management.
Cloud Native: 10 Years and Beyond
Experts from Akamai and Microsoft spoke about the past decade of cloud-native development and its outlook for the future.
They reported that more than 70 CNCF projects are growing healthily, with efforts accelerating particularly in the fields of AI and security.
A speaker from Amazon also gave a concrete example of how they use machine learning to plan service capacity and measure risk during large-scale events like Black Friday.
They also discussed the importance of team collaboration across all services and stress testing based on dynamic forecasts in the event of sudden increases in traffic that simple autoscaling alone cannot handle.
Particularly interesting sessions
Here are three sessions I found particularly interesting.
In-Place Pod Resize in Kubernetes: Dynamic Resource Management Without Restarts
resizePolicySecuring Data Applications at Pinterest With Finer Grained Access Control on Kubernetes
valanceTurbocharging Argo CD: Replacing Redis With Dragonfly for Better Performance and Lower Bills
An engineer from Acuity, who also maintains Argo CD, presented a case study in which they replaced Redis, their cache, with Dragonfly to improve Argo CD's performance and costs.
Argo CD uses Redis as a cache to improve UI response speed, among other things. However, configuring it for high availability (HA) requires a large number of pods and containers, posing a challenge in terms of resource consumption.
By replacing it with Dragonfly, a new in-memory cache that is compatible with Redis, they reported being able to reduce the number of pods by approximately 30% and the number of containers by approximately 40%.
In one customer case, memory utilization dropped dramatically from 9% to 3%, resulting in cost savings.
Dragonfly has already reached general availability (GA), and the presenter emphasized that it is "production ready" and "in use in production environments."
As we use Argo CD in many of our services, this case study provided extremely useful information that we could put to use immediately.
summary
Participating in the four-day KubeCon was a great opportunity to absorb new technology trends.
In particular, I realized that themes such as platform engineering, AI infrastructure, and the security and cost optimization that support them will be important guidelines for future technology selection and system design.
Furthermore, the enthusiasm expressed in each session and the opportunity to interact with engineers from all over the world greatly increased my motivation for technology.
I would like to use the knowledge and perspective I gained here to identify issues facing my company's services and use them to come up with new proposals to address them.
SRG is looking for people to work with us.
If you are interested, please contact us here.