AWS Cost Optimization: Reserved Instances, Savings Plans, S3 Lifecycle, and RDS Right-Sizing

AWS bills are often 40–60% higher than they need to be. The excess comes from on-demand pricing for workloads that run continuously, oversized instances, S3 storage that never gets accessed, and data transfer charges that no one noticed.

This guide covers the changes that have the highest return per hour of engineering time.

---

Where the Money Goes

For a typical SaaS product, cost typically breaks down:

---

Commitment Discounts: Savings Plans vs Reserved Instances

Which to choose:

• Compute Savings Plan: Most flexible — covers any EC2, Fargate, or Lambda in any region. Choose this for most teams.
• EC2 Instance Savings Plan: Covers specific instance family in one region (e.g., m5 in us-east-1). Higher discount (~72%) but less flexible.
• Reserved Instances: Still worth it for RDS and ElastiCache (Savings Plans don't cover these).

How to buy Savings Plans:

# Step 1: Check your usage with AWS Cost Explorer
# Look at: Cost Explorer → Savings Plans → Recommendations
# AWS calculates the optimal commitment based on your last 7/30/60 days

# Step 2: Calculate commitment amount
# Don't commit to 100% of baseline — leave room for variability
# Typical: commit to 70–80% of steady-state usage

# Step 3: Purchase via Console or CLI
aws savingsplans purchase-savings-plan \
  --savings-plan-offering-id <id-from-recommendations> \
  --commitment 0.50 \  # $0.50/hour commitment
  --purchase-time 2026-06-01T00:00:00Z

Savings example:

• On-demand: 4× m5.xlarge ECS Fargate tasks, running 24/7 = ~$480/month
• Compute Savings Plan (1-year, no upfront): ~$320/month (33% savings)
• Compute Savings Plan (1-year, all upfront): ~$280/month (42% savings)

---

RDS Cost Optimization

RDS is often the second-largest line item and has multiple optimization levers:

Right-Sizing

# Find oversized RDS instances (CloudWatch metrics)
aws cloudwatch get-metric-statistics \
  --namespace AWS/RDS \
  --metric-name CPUUtilization \
  --dimensions Name=DBInstanceIdentifier,Value=production-postgres \
  --start-time 2026-04-01T00:00:00Z \
  --end-time 2026-05-01T00:00:00Z \
  --period 86400 \
  --statistics Average Maximum

# If CPU avg < 10% and max < 40%: you're probably overprovisioned
# Check also: FreeableMemory, ReadIOPS, WriteIOPS

Instance family stepping down (example):

db.r6g.4xlarge (128GB RAM, $1,400/mo) → db.r6g.2xlarge (64GB RAM, $730/mo)
Justification: FreeableMemory never drops below 30GB
Savings: $670/month ($8,040/year)

Reserved Instances for RDS

# Purchase RDS Reserved Instance
aws rds purchase-reserved-db-instances-offering \
  --reserved-db-instances-offering-id <offering-id> \
  --reserved-db-instance-id production-postgres-ri \
  --db-instance-count 1
# Typical savings: 36–42% vs on-demand for 1-year no-upfront

Aurora Serverless v2 for Variable Workloads

# terraform/rds.tf
resource "aws_rds_cluster" "aurora_serverless" {
  cluster_identifier      = "api-db"
  engine                  = "aurora-postgresql"
  engine_mode             = "provisioned"
  engine_version          = "15.4"
  database_name           = "app"
  master_username         = "admin"
  master_password         = var.db_password

  serverlessv2_scaling_configuration {
    min_capacity = 0.5   # 0.5 ACUs minimum (~$0.06/hr at minimum)
    max_capacity = 4     # Scale up to 4 ACUs under load
  }
}

resource "aws_rds_cluster_instance" "aurora_serverless" {
  cluster_identifier = aws_rds_cluster.aurora_serverless.id
  instance_class     = "db.serverless"
  engine             = aws_rds_cluster.aurora_serverless.engine
  engine_version     = aws_rds_cluster.aurora_serverless.engine_version
}
# Cost: $0.12/ACU-hour. At 0.5 ACU idle: ~$43/month
# vs db.t3.medium always-on: ~$60/month with more disk overhead
# Better for dev/staging with variable load

---

S3 Cost Optimization

S3 has multiple storage classes. Most teams keep everything in Standard when they should be using tiered storage:

Lifecycle policy:

{
  "Rules": [
    {
      "ID": "IntelligentTieringForUserUploads",
      "Status": "Enabled",
      "Filter": { "Prefix": "uploads/" },
      "Transitions": [
        {
          "Days": 30,
          "StorageClass": "INTELLIGENT_TIERING"
        }
      ]
    },
    {
      "ID": "ArchiveOldLogs",
      "Status": "Enabled",
      "Filter": { "Prefix": "logs/" },
      "Transitions": [
        { "Days": 30, "StorageClass": "STANDARD_IA" },
        { "Days": 90, "StorageClass": "GLACIER_INSTANT_RETRIEVAL" },
        { "Days": 365, "StorageClass": "DEEP_ARCHIVE" }
      ],
      "Expiration": { "Days": 2555 }  // Delete after 7 years
    },
    {
      "ID": "DeleteIncompleteMultipartUploads",
      "Status": "Enabled",
      "Filter": {},
      "AbortIncompleteMultipartUpload": { "DaysAfterInitiation": 7 }
    }
  ]
}

# Apply lifecycle policy
aws s3api put-bucket-lifecycle-configuration \
  --bucket your-bucket \
  --lifecycle-configuration file://lifecycle.json

---

ECS Fargate Spot

Fargate Spot uses spare AWS capacity at 70% discount. Suitable for fault-tolerant, interruptible workloads:

# terraform/ecs.tf
resource "aws_ecs_service" "worker" {
  name            = "background-worker"
  cluster         = aws_ecs_cluster.main.id
  task_definition = aws_ecs_task_definition.worker.arn
  desired_count   = 5

  capacity_provider_strategy {
    capacity_provider = "FARGATE"
    weight            = 1
    base              = 1  # At least 1 on-demand task (safety)
  }

  capacity_provider_strategy {
    capacity_provider = "FARGATE_SPOT"
    weight            = 4  # 4× more Spot than on-demand (80% Spot)
  }
}

Don't use Spot for your API service — interruptions cause 2-minute drains during which capacity drops. Use Spot for: workers, batch jobs, CI builds, data processing.

---

Data Transfer Cost Reduction

Data transfer between AWS services in the same region is free. Cross-region and internet egress is expensive:

# Common data transfer charges to eliminate:

# 1. Use VPC endpoints for S3/DynamoDB (free transfer within VPC)
aws ec2 create-vpc-endpoint \
  --vpc-id vpc-xxx \
  --service-name com.amazonaws.us-east-1.s3 \
  --route-table-ids rtb-xxx
# Saves: $0.01/GB for S3 traffic that previously went through NAT gateway

# 2. Put CloudFront in front of S3 (CloudFront egress is cheaper than S3 direct)
# S3 → internet: $0.09/GB
# CloudFront → internet: $0.0085/GB (90% cheaper for high volume)

# 3. Keep services in the same AZ when possible
# Cross-AZ data transfer: $0.01/GB each way
# Same-AZ: free

---

Cost Monitoring: AWS Cost Anomaly Detection

# Set up cost anomaly detection via AWS CLI
aws ce create-anomaly-monitor \
  --anomaly-monitor '{
    "MonitorName": "Service Monitor",
    "MonitorType": "DIMENSIONAL",
    "MonitorDimension": "SERVICE"
  }'

# Subscribe to alerts (email when anomaly detected)
aws ce create-anomaly-subscription \
  --anomaly-subscription '{
    "SubscriptionName": "DailyAnomalyAlert",
    "MonitorArnList": ["arn:aws:ce::xxx:anomalymonitor/xxx"],
    "Subscribers": [
      { "Address": "engineering@yourcompany.com", "Type": "EMAIL" }
    ],
    "Threshold": 20,
    "Frequency": "DAILY"
  }'

Terraform for budget alerts:

resource "aws_budgets_budget" "monthly" {
  name         = "Monthly AWS Budget"
  budget_type  = "COST"
  limit_amount = "5000"
  limit_unit   = "USD"
  time_unit    = "MONTHLY"

  notification {
    comparison_operator        = "GREATER_THAN"
    threshold                  = 80
    threshold_type             = "PERCENTAGE"
    notification_type          = "ACTUAL"
    subscriber_email_addresses = ["engineering@yourcompany.com"]
  }

  notification {
    comparison_operator        = "GREATER_THAN"
    threshold                  = 100
    threshold_type             = "PERCENTAGE"
    notification_type          = "FORECASTED"
    subscriber_email_addresses = ["engineering@yourcompany.com", "cto@yourcompany.com"]
  }
}

---

Working With Viprasol

We conduct AWS cost optimization audits — identifying savings opportunities, purchasing commitment discounts, implementing S3 lifecycle policies, right-sizing databases, and setting up cost monitoring. Most audits find 30–50% in immediate savings.

→ Talk to our cloud team about AWS cost optimization.

---

See Also

• Infrastructure as Code — Terraform for reproducible, cost-controlled infra
• Serverless Cost Optimization — Lambda-specific cost reduction
• Kubernetes vs ECS — choosing compute that fits your cost model
• DevOps Best Practices — cost monitoring in the engineering process
• Cloud Solutions — cloud infrastructure and cost management