Cloud Cost Optimization

Overview

Cloud costs can spiral unexpectedly — unused resources, oversized instances, and forgotten development environments silently drain budgets. Cost optimization isn’t just about cutting spending; it’s about aligning infrastructure capacity with actual demand. This resource covers right-sizing, purchasing strategies (reserved vs. spot), automated scheduling, and FinOps practices that reduce waste without impacting reliability.

When to Use

Use this resource when:

• Monthly cloud bills are growing faster than user traffic
• Development and staging environments run 24/7 despite only being used during business hours
• You’re paying for overprovisioned instances that use <20% CPU
• You need to justify infrastructure costs to finance or leadership

Solution

AWS Cost Explorer Analysis (AWS CLI)

# Find top cost drivers by service
aws ce get-cost-and-usage \
  --time-period Start=$(date -d '30 days ago' +%Y-%m-%d),End=$(date +%Y-%m-%d) \
  --granularity MONTHLY \
  --metrics BlendedCost \
  --group-by Type=DIMENSION,Key=SERVICE \
  --query 'ResultsByTime[0].Groups[?Metrics.BlendedCost.Amount > \`100\`].Keys'

# Find unattached EBS volumes
aws ec2 describe-volumes \
  --filters Name=status,Values=available \
  --query 'Volumes[*].[VolumeId,Size,CreateTime]'

Terraform Scheduled Scaling

resource "aws_autoscaling_schedule" "dev_office_hours" {
  scheduled_action_name  = "dev-office-hours"
  min_size               = 1
  max_size               = 3
  desired_capacity       = 2
  recurrence             = "0 9 * * MON-FRI"  # 9 AM UTC
  autoscaling_group_name = aws_autoscaling_group.dev.name
}

resource "aws_autoscaling_schedule" "dev_night_shutdown" {
  scheduled_action_name  = "dev-night-shutdown"
  min_size               = 0
  max_size               = 0
  desired_capacity       = 0
  recurrence             = "0 18 * * MON-FRI" # 6 PM UTC
  autoscaling_group_name = aws_autoscaling_group.dev.name
}

Spot Instance with Fallback (Kubernetes)

apiVersion: apps/v1
kind: Deployment
metadata:
  name: spot-workload
spec:
  replicas: 5
  template:
    spec:
      affinity:
        nodeAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
            - weight: 100
              preference:
                matchExpressions:
                  - key: node-type
                    operator: In
                    values: [spot]
      tolerations:
        - key: spot
          operator: Equal
          value: "true"
          effect: NoSchedule
      containers:
        - name: app
          image: myapp:latest

Explanation

Four pillars of cloud cost optimization:

1. Right-size: Match instance type to actual usage; downsize overprovisioned resources
2. Reserved capacity: Commit to 1-3 year reserved instances for predictable workloads (40-60% savings)
3. Spot/preemptible: Use interruptible instances for fault-tolerant batch jobs (60-90% savings)
4. Auto-scheduling: Turn off dev/staging environments nights and weekends

FinOps lifecycle:

• Inform: Visibility into cloud spend per team, project, and environment
• Optimize: Technical and rate optimizations (RI, spot, rightsizing)
• Operate: Continuous governance, budgets, and automated policies

Variants

Strategy	Savings	Effort	Risk
Reserved instances	40-60%	Low	Commitment lock-in
Spot instances	60-90%	Medium	Interruption
Scheduled shutdown	50-70%	Low	Manual oversight
Storage tiering	30-50%	Low	Access latency
Serverless	Variable	Medium	Cold start

Best Practices

• Tag everything: Cost allocation tags (team, project, environment) enable chargeback
• Set budgets and alerts: Alert at 80% of monthly budget; investigate immediately
• Review unused resources weekly: Dangling IPs, orphaned volumes, and stale snapshots add up
• Use Savings Plans over RIs: More flexible; apply across instance families and regions
• Implement auto-scaling: Scale to zero for dev environments; scale up for production peaks. See autoscaling policies.

Common Mistakes

1. No cost ownership: When engineering doesn’t see the bill, waste accumulates
2. Overcommitting to reserved instances: Buying 3-year RIs for workloads that may migrate to serverless
3. Ignoring data transfer costs: NAT Gateway, cross-AZ traffic, and egress can exceed compute costs
4. Leaving preview resources running: POCs and experiments that become permanent line items
5. One-size-fits-all pricing: Production needs stability; dev can tolerate spot interruptions

Frequently Asked Questions

Q: Should I use spot instances for production? A: Only for stateless, fault-tolerant workloads with proper fallback to on-demand. Never for databases or singleton services.

Q: How do I prevent developers from creating expensive resources? A: SCPs (Service Control Policies) restrict instance types by OU. Terraform policies enforce approved instance families.

Q: What’s the difference between FinOps and DevOps? A: DevOps optimizes for speed and reliability. FinOps adds cost as a first-class metric, with cross-functional accountability.