cost-optimization

Container Right-Sizing

Monitor CPU and memory for 7 days before setting limits. Set requests to P50, limits to P99.

BAD: Guessing resource limits. Over-provisioned containers waste money.

resources:
  requests:
    memory: "1Gi"
    cpu: "1000m"
  limits:
    memory: "2Gi"
    cpu: "2000m"

GOOD: Use actual usage data from monitoring.

kubectl top pod myapp-12345 --containers

If P50 is 200Mi/100m and P99 is 400Mi/300m:

resources:
  requests:
    memory: "200Mi"
    cpu: "100m"
  limits:
    memory: "400Mi"
    cpu: "300m"

Vertical Pod Autoscaler

Get sizing recommendations from real workload data.

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: myapp-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: myapp
  updateMode: "Off"  # Recommendation mode only

Check recommendations:

kubectl describe vpa myapp-vpa

Cost Calculation

Example: 10 pods running 24/7 with 1Gi memory, 1 CPU.

• Memory: 10 pods x 1 GiB x $0.0004/hr = $29/month
• CPU: 10 pods x 1 core x $0.04/hr = $292/month
• Total: $321/month

Right-size to 200Mi memory, 100m CPU:

• Memory: 10 pods x 0.2 GiB x $0.0004/hr = $5.80/month
• CPU: 10 pods x 0.1 core x $0.04/hr = $29.20/month
• Total: $35/month

Savings: 89% ($286/month)

Horizontal Pod Autoscaler

Scale replicas based on actual load. Do not run excess capacity during low traffic.

BAD: Fixed replica count. Wastes money overnight and weekends.

spec:
  replicas: 10

GOOD: HPA scales from 2 to 10 based on CPU usage.

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: myapp-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: myapp
  minReplicas: 2
  maxReplicas: 10
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 70

Cost Impact