Perimattic

Kubernetes Cost Optimizer

Optimize your Kubernetes spending. Input your cluster configuration to find over-provisioned resources, spot instance opportunities, and potential savings.

  • Free Optimization
  • Cluster Analysis
  • Cost Breakdown
Cloud-native infrastructure and containerized services running at scale
ISO 27001 Certified
AWS Partner
Microsoft Partner
ServiceNow Partner
Calculator

Kubernetes Cost Optimizer

Enter your details below to get your personalised analysis
How it works

How does the Kubernetes cost optimizer work?

It compares what your cluster costs today with a right-sized cluster that moves suitable workloads to spot nodes.

Most clusters are sized for peak load and then left alone, so nodes sit at 30–40% CPU for most of the month. The optimizer starts from your node count, the vCPUs and memory on each node, and your average CPU and memory utilization. From those it works out how much capacity your workloads actually use.

It then sizes a new cluster so that the busier of the two resources, CPU or memory, runs at about 70%. That leaves headroom for traffic spikes, rolling deployments and node failures while removing nodes you pay for but do not use. Sizing on whichever resource is tighter stops the tool from suggesting a cluster that looks cheap on CPU but runs out of memory.

Finally, it moves the share of workloads you mark as spot-eligible onto spot or preemptible nodes, using an average 65% discount on the on-demand price. The rest stay on on-demand nodes. The result shows your current monthly cost, the optimized cost, the monthly saving and the saving as a percentage.

Inputs and what they mean

Nodes, vCPUs and memory per node
Your current cluster capacity. Count worker nodes only, not the control plane.
Average CPU and memory utilization
A 30-day average from your monitoring tool (Prometheus, Datadog, CloudWatch or your cloud console), not a single snapshot.
On-demand cost per node
The monthly price of one node from your cloud bill or the provider's price list.
Spot-eligible workloads
The share of pods that can be restarted without harm, such as stateless services, batch jobs and CI runners.
FAQ

Frequently asked questions

How much can I save by right-sizing Kubernetes clusters?

Most organizations are 40-60% over-provisioned. Right-sizing typically saves 30-40%, and adding spot instances can bring total savings to 50-70%.

What workloads are suitable for spot instances?

Stateless microservices, batch processing, CI/CD runners, development/staging environments, and any workload that can tolerate interruption with graceful shutdown.

How often should I review Kubernetes resource allocation?

Review monthly at minimum. Use tools like Kubecost, OpenCost, or cloud-native recommendations (GKE Autopilot, EKS right-sizing) for continuous optimization.

What is a good CPU utilization target for Kubernetes?

Target 65-75% average CPU utilization for production workloads. Below 40% indicates significant over-provisioning. Above 85% risks performance degradation during spikes.

Why does the optimizer size the cluster to 70% utilization and not 100%?

Running nodes close to full leaves no room for traffic spikes, rolling deployments or a node failing. About 70% on the tighter resource, CPU or memory, is a common balance between cost and reliability. If your workloads are very spiky, aim lower.

Where do I find my cluster's average CPU and memory utilization?

Use a 30-day average from your monitoring stack: Prometheus with kube-state-metrics, Datadog, CloudWatch Container Insights, Azure Monitor or GKE's built-in metrics. A snapshot taken at one point in the day is misleading.

Does the result include savings from autoscaling?

No. The estimate covers right-sizing and spot nodes only. Cluster and pod autoscaling can save more on top, because they shrink the cluster at night and at weekends when load drops.

Are spot instances safe for production workloads?

They are safe for workloads that tolerate interruption. Run several replicas across node pools, set pod disruption budgets, handle the termination notice gracefully, and keep databases and other stateful services on on-demand nodes.

DevOps & Platform Engineering

Overpaying for Kubernetes?

Most teams overspend by 50%+ on Kubernetes. We help you right-size clusters, implement spot strategies, and set up cost governance — with zero downtime.

  • Free consultation
  • No commitment
  • Tailored to your stack