Kuberneteshe Container Management Orchestration Systemhas proven vital to modern computing, except in one area: artificial intelligence (AI) and machine learning (ML). The problem: AI and ML require substantial CPU, memory, and GPU resources, which are Not easy to manage in Kubernetes.
Now with the latest version of Kubernetes: Kubernetes 1.31, Elli — he Cloud Native Computing Foundation (CNCF) is addressing these issues.
Also: Five ways CIOs can manage enterprise demand for generative AI
Elli’s enhanced AI features begin with alpha support for Open Container Initiative (OCI) images and artifacts as a native volume source. This may not sound like much, but it allows developers to switch large language models (LLMs) as easily as they do with regular container images.
Elli also brings an updated API and dynamic resource allocation design to Kubernetes. This feature will help standardize access and management of hardware accelerators, such as GPUs, that are essential for AI and ML. It also simplifies the implementation of features such as cluster auto-scaling, which in turn will make it easier to run AI and ML jobs on Kubernetes.
In the past, Kubernetes had multiple ways to access a host’s underlying hardware. The updated Dynamic Resource Allocation (DRA) provides a simpler way to access system resources. The old style, using a DRA controller, is still supported via “Classic DRA.”
Leaving AI behind, Kubernetes now also fully supports Application Armora Linux kernel security module that allows system administrators to restrict program capabilities with per-program profiles. This feature is now generally available and allows users to configure AppArmor profiles for containers directly through the Kubernetes API. If implemented properly, AppArmor support will help make Kubernetes clusters and workloads more secure.
On the security front, a new optional feature allows administrators to configure endpoints so that anonymous access requests can be blocked. This will help protect clusters from role-based access control (RBAC) misconfigurations that could otherwise grant anonymous users broad access to the cluster.
Also: OIN extends Linux patent protection once again (but not to AI)
As an ever-evolving open source program, Kubernetes 1.31 also continues to optimize and modernize its codebase by removing deprecated features. This includes:
- Removing cloud provider integrations within the tree: Completing a process that began with release 1.26, Kubernetes 1.31 removes all cloud provider integrations within the tree. This change is part of an ongoing effort to ensure that Kubernetes remains a vendor-agnostic platform. You can still integrate your Kubernetes clusters with a specific cloud provider. However, you won’t be able to do so from within Kubernetes. You’ll need to use the recommended approach via external integration.
- Deprecation of non-CSI volume limit plugins: This release deprecates all non-CSI volume limit scheduler plugins, in line with Kubernetes’ strategy to transition to the Container Storage Interface (CSI) for all storage-related functionality.
- Disuse of Group C version 1: Cgroups is a Linux kernel feature that allows you to allocate, prioritize, and manage system resources between processes. From here, groups v2 This is what you should be using.
Putting it all together, this release goes far beyond making Kubernetes AI-ready. Angelos Kolaitis, a Canonical Senior software engineer and team leader for Kubernetes version 1.31 said this new release is about “taking control.” complexity and the specific details of implementing Kubernetes code“make Kubernetes focus on the desired state and leave all deployments and all additional source code out of it.”
Also: What is Kubernetes and why is it so important?
Beyond the new features, Kolaitis said, “You don’t have to do anything, but Kubernetes will take additional steps to ensure your workloads are running, are reliable, and your production services can be counted on to stay up and running.”
New features are nice, but stability is what matters most. For that reason alone, I recommend exploring an upgrade to Kubernetes 1.31 as soon as possible.