Red Hat has released a new version of its flagship platform, OpenShift 4.21. The main focus of this release is on optimizing workflows for artificial intelligence (AI) and machine learning (ML).
OpenShift is a platform for managing containers and applications built on Kubernetes. Simply put, it helps developers launch and scale software products in the cloud or on corporate servers without getting bogged down in infrastructure configuration complexities. In recent years, companies have increasingly used such solutions to train and deploy neural networks – and this is precisely the demand the fresh update addresses.
New Features for AI and Machine Learning Workloads
What's Changed in Handling AI Workloads
One of the key innovations is the integration of the Kueue tool, version 1.2. This is a system for managing compute job queues. When multiple model training or data processing tasks are running simultaneously, Kueue distributes resources to prevent stalls and ensure all operations are executed according to priority.
For those working with distributed computing – for instance, when a model is trained across multiple nodes or GPUs – JobSet support has been added. This mechanism allows you to run groups of related tasks as a single unit. In short, instead of manually monitoring the synchronous start and completion of every process stage, JobSet automates the cycle.
Dynamic GPU Resource Allocation in OpenShift 4.21
Smart GPU Allocation
Another important feature is Dynamic Resource Allocation. In the context of machine learning, this primarily concerns graphics processing units (GPUs), which are the main «engine» for training models.
Usually, GPUs are assigned to tasks statically. If a process finishes faster than expected or, conversely, requires more power, it results in either hardware downtime or performance «bottlenecks». Dynamic allocation allows GPU capacity to be redistributed «on the fly» depending on actual load. This is critically important when compute resources are expensive and need to be used with maximum efficiency.
Benefits of OpenShift 4.21 for AI Development
Why It Matters
All these changes aim for one goal: to make launching AI projects simpler and cheaper. Model training always involves huge volumes of data, long execution times, and expensive hardware. The more efficiently the platform manages the workload, the less time is wasted on waiting and the lower the infrastructure maintenance costs.
OpenShift 4.21 doesn't «reinvent the wheel» – most of these functions were already present in the Kubernetes ecosystem. However, their seamless integration into a single platform with an intuitive interface and official Red Hat support makes the technology more accessible to companies not ready to build complex infrastructure from scratch themselves.
Target Audience and Use Cases for the New Release
Who Is This For
Primarily, this update will be appreciated by Data Science teams in large organizations where regular model training, running multiple parallel experiments, and sharing limited capacity across different departments is required.
For individual developers or small startups, these capabilities might prove excessive – often it's simpler for them to rent a GPU in the cloud directly. But if a company has its own infrastructure or adheres to strict data security requirements, OpenShift 4.21 becomes a rational choice.
The new release is a consistent step toward ensuring corporate tools keep pace with AI development. It is difficult to judge yet how widespread the adoption of these new functions will be, but the direction chosen is correct: less manual administration, more automation, and maximum efficiency from using the «hardware».