"Red Hat OpenShift 4.21 is now generally available, based on Kubernetes 1.34 and CRI-O 1.34. This release focuses on unifying AI training, containerized microservices, and virtualized applications on a single operational model to reduce costs and eliminate infrastructure silos. Accelerated AI innovation Streamlined training: Data scientists can use a single TrainJob API instead of managing separate machine learning frameworks. Queue visibility: A new Visibility API for pending workloads provides estimated start times and helps administrators identify resource bottlenecks."
"Intelligent GPU allocation: New Dynamic Resource Allocation (DRA) capabilities allow teams to request GPUs by specific attributes (e.g., vRAM size) and define fallback strategies to maximize hardware utilization. Core infrastructure and cost efficiency Scaling to zero: Hosted control planes and NodePools can now hibernate during inactivity and resume automatically, eliminating costs for idle infrastructure. Dynamic right-sizing: Integration with the VerticalPodAutoscaler (VPA) allows control planes to scale based on real-time memory consumption rather than static estimates."
OpenShift 4.21 is based on Kubernetes 1.34 and CRI-O 1.34 and unifies AI training, containerized microservices, and virtualized applications on a single operational model to reduce costs and remove silos. Data scientists can use a single TrainJob API for training, and a Visibility API provides estimated start times for pending workloads to highlight bottlenecks. The JobSet Operator enables orchestration of interdependent jobs, while Dynamic Resource Allocation lets teams request GPUs by attributes and define fallbacks. Hosted control planes and NodePools can hibernate and resume to eliminate idle costs. Integration with VPA enables memory-based scaling, and ecosystem and virtualization enhancements expand platform flexibility.
Read at App Developer Magazine
Unable to calculate read time
Collection
[
|
...
]