EKS Cost Optimization

Optimization Strategies

1. Right-size Resources

# Check actual vs requested resources
kubectl top pods -A --containers=true | head -20
 
# Right-size container requests
resources:
  requests:
    memory: "256Mi"
    cpu: "250m"
  limits:
    memory: "512Mi"
    cpu: "500m"

2. Use Spot Instances

# Node group with mixed instances
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig
metadata:
  name: my-cluster
managedNodeGroups:
  - name: spot
    instanceTypes: [t3.medium, t3a.medium, m5.large]
    spot: {}
    labels:
      lifecycle: Ec2Spot
    taints:
      - key: "spot"
        value: "true"
        effect: "NoSchedule"

3. Graviton for Compute

# Migrate to Graviton
eksctl create nodegroup \
  --cluster my-cluster \
  --name graviton \
  --node-type m6g.xlarge \
  --nodes 3 \
  --managed

4. Use Karpenter for Right-sizing

apiVersion: karpenter.sh/v1
kind: NodePool
metadata:
  name: default
spec:
  template:
    spec:
      requirements:
        - key: karpenter.sh/capacity-type
          operator: In
          values: [spot]
        - key: karpenter.sh/provisioner-name
          operator: In
          values: [default]
      limits:
        cpu: 100
        memory: 100Gi
      consolidation:
        enabled: true

5. Storage Optimization

Strategy	Savings
Use gp3 instead of gp2	~20% cheaper
Delete unused PVCs	$50-200/month
Use S3 for object storage	vs EBS for large data

6. Cluster Autoscaler

# Ensure proper resource requests
# HPA with scale-down enabled
behavior:
  scaleDown:
    stabilizationWindowSeconds: 300
    policies:
    - type: Percent
      value: 10
      periodSeconds: 60

7. Reserved Capacity / Savings Plans

Option	Savings	Flexibility
On-Demand	Baseline	Highest
1yr Reserved	~30%	Medium
3yr Reserved	~60%	Low
Savings Plans	~30-60%	High

8. kubecost Monitoring

helm install kubecost kubecost/cost-analyzer \
  --namespace kubecost \
  --create-namespace

Cost Allocation Tags

# Enable cost allocation in AWS
aws ce enable-split-cost-allocation-resources

Quick Wins Checklist

• Set resource requests/limits on all pods
• Use Spot for non-critical workloads
• Use Graviton for Linux workloads
• Enable Karpenter or Cluster Autoscaler
• Delete unused namespaces/deployments
• Use gp3 storage instead of gp2
• Review Kubecost recommendations weekly

References

• EKS Cost Optimization
• Kubecost
• EKS Workshop - Cost