Cut Costs and Scale Smarter on ROSA with Karpenter on Amazon Web Services

Organizations running containerized applications on Red Hat OpenShift Service on AWS (ROSA) benefit from Kubernetes elasticity, but many still experience resource underutilization due to rigid node groups and manual scaling practices.

To address this, Red Hat offers the Red Hat build of Karpenter (Technology Preview), a fully managed autoscaler based on the AWS-created open-source project. Karpenter transforms how compute capacity is provisioned by launching the right Amazon EC2 instances exactly when workloads need them—no predefined instance types, no over-provisioned pools.

With guidance from Softprom, official AWS Partner, organizations can adopt Karpenter to significantly lower ROSA compute costs while improving operational efficiency.

The node scaling challenge in ROSA

Traditional approaches rely on:

• Static machine pools with fixed instance types
• Kubernetes Cluster Autoscaler reacting by adding more of the same nodes
• Over-provisioning to avoid capacity shortages
• Poor bin-packing and stranded resources
• Manual management of multiple node pools
• Availability issues when a specific instance type is constrained

At scale, these inefficiencies multiply across environments and teams, driving unnecessary cloud spend.

How Karpenter changes the model

Unlike node-group based autoscalers, Karpenter works at the pod level.

When pods cannot be scheduled, Karpenter:

1. Analyzes CPU, memory, GPU, topology, and affinity requirements
2. Simulates scheduling across hundreds of instance options
3. Uses the EC2 CreateFleet API to provision the optimal instance
4. Continuously evaluates nodes for consolidation and right-sizing

Karpenter can select from 400+ EC2 instance types across:

• x86 and Arm (AWS Graviton) architectures
• On-Demand and Spot markets
• Multiple Availability Zones

The result is infrastructure that is always right-sized and continuously optimized.

Cost optimization mechanisms

Karpenter delivers savings through:

Vertical right-sizing

Instance types and sizes adapt dynamically to workload demand.

Horizontal elasticity

Worker nodes scale up and down precisely with application needs.

Market-aware provisioning

Automatic use of Spot, On-Demand Capacity Reservations (ODCRs), and On-Demand based on availability and price.

Continuous consolidation

Underutilized nodes are replaced with better-suited instances.

Operational savings

Platform teams no longer manage complex machine pool logic.

Benefits of Red Hat build of Karpenter for ROSA (HCP)

When used with ROSA hosted control planes:

• Karpenter controllers run in the control plane (no worker overhead)
• Can be enabled on existing clusters after upgrade
• Independent upgrades of control plane, nodes, and Karpenter
• Can coexist with Cluster Autoscaler for gradual migration
• Supports ODCRs and EC2 Capacity Blocks for ML workloads
• Allows advanced kubelet tuning and node optimization

Sustainability and price-performance

Karpenter intelligently prioritizes AWS Graviton instances where suitable, improving:

• Price-performance ratio
• Energy efficiency
• Alignment with the AWS Well-Architected Sustainability Pillar

Security and compliance alignment

Karpenter integrates natively with enterprise security models:

• Uses OpenShift-approved AMIs
• Launches nodes only in approved VPC subnets and security groups
• Integrates with AWS Identity and Access Management and AWS STS for least-privilege access
• Supports Red Hat compliance standards (FIPS, SOC2, FedRAMP)
• Respects Kubernetes taints, tolerations, and scheduling constraints

Practical outcomes for ROSA customers

Organizations adopting Karpenter on ROSA typically achieve:

• Significant reduction in over-provisioned compute
• Better bin-packing and node utilization
• Faster workload scheduling
• Lower operational overhead for platform teams
• Improved reliability during capacity shortages

How Softprom helps

Softprom supports ROSA customers with:

• Assessment of current node utilization and cost profile
• Karpenter architecture and migration planning
• Integration with existing autoscaling and governance policies
• Optimization for Spot, Graviton, and capacity reservations
• Ongoing AWS cost optimization strategy

Conclusion

Karpenter fundamentally changes compute provisioning in ROSA. By moving from static node groups to intelligent, pod-aware autoscaling, organizations reduce waste, improve utilization, and simplify operations. As the Red Hat build of Karpenter becomes generally available for ROSA, this capability will become a key lever for OpenShift cost optimization on AWS.