AWS Cost Optimization for Startups: Real Strategies That Cut Our Bill by 53%

Consolidated from real AWS cost analysis, auditing, and optimization sessions. All account-specific identifiers replaced with placeholders.

Cost Focus 1: How to keep this maintainable at scale for predictable operations (Aws Cost Optimization)

This article documents practical AWS cost optimization strategies applied to a production web application running on ECS Fargate, ALB, and RDS PostgreSQL — taking projected monthly costs from ~$183/month down to ~$85/month through methodical analysis and incremental changes.

---

Editorial review note for Aws Cost Optimization #73

This section was reviewed by a human editor to keep the recommendations actionable and technically grounded. Reviewed by: Med Amine Mahmoud. Last editorial review: 2026-05-26T16:10:01Z.

Cost Focus 3: Risk controls worth enforcing early for cleaner ownership (Aws Cost Optimization)

Default retention: forever (costs grow indefinitely).

# Set 7-day retention on dev, 30-day on prod
aws logs put-retention-policy --log-group-name /ecs/my-app-dev --retention-in-days 3
aws logs put-retention-policy --log-group-name /ecs/my-app-prod --retention-in-days 30

---

Cost Focus 4: Signals that tell you this is working for measurable outcomes (Aws Cost Optimization)

---

Cost Focus 5: How to keep cost and reliability aligned for fewer incident surprises (Aws Cost Optimization)

Before vs After

---

Cost Focus 6: What to document for your team for this workload (Aws Cost Optimization)

1. Measure before optimizing — Use Cost Explorer and CloudWatch to identify actual usage patterns
2. Right-size aggressively — Most dev workloads are massively over-provisioned
3. Eliminate idle resources — Auto-shutdown for non-production, release unattached IPs
4. Consolidate where possible — One ALB with routing rules beats multiple ALBs
5. Set lifecycle policies — Prevent unbounded growth in ECR, CloudWatch Logs, and S3
6. Re-evaluate architecture — NAT Gateways and other "best practice" resources may not be needed for your workload

Cost Focus 7: Where this architecture earns its value for your runbook (Aws Cost Optimization)

Resource Inventory & Monthly Costs

After deploying to AWS, the first step was a comprehensive resource audit:

# Discover all billable resources
aws ecs list-services --cluster my-cluster
aws rds describe-db-instances
aws elbv2 describe-load-balancers
aws ec2 describe-addresses # Elastic IPs
aws ecr describe-repositories

Initial Cost Breakdown

---

Cost Focus 8: Operational notes from real-world usage for production readiness (Aws Cost Optimization)

Discovery

CloudWatch Container Insights revealed:

• Average CPU utilization: 8-12% (on 2 vCPU)
• Average memory utilization: 35% (on 4 GB)
• Peak CPU: 45% (during deployments only)

Action

Downgraded task definition from 2 vCPU / 4 GB to 1 vCPU / 2 GB:

aws ecs register-task-definition `
--family my-app `
--cpu 1024 `
--memory 2048 `
--container-definitions file://task-def.json

Result

• Savings: ~$36/month (50% reduction in Fargate compute)
• No performance impact at current traffic levels
• Auto-scaling still active (1→8 tasks) for spike protection

---

Cost Focus 9: How to avoid expensive rework for sustained reliability (Aws Cost Optimization)

Problem

Two separate ALBs — one for production, one for dev — at $16/month each.

Solution

Consolidated to a single ALB with host-header routing:

# Add dev rule to prod ALB HTTPS listener
aws elbv2 create-rule `
--listener-arn $HTTPS_LISTENER_ARN `
--priority 10 `
--conditions "Field=host-header,Values=[dev.example.com]" "Field=source-ip,Values=[YOUR_IP/32]" `
--actions "Type=forward,TargetGroupArn=$DEV_TG_ARN"

Architecture After

Single ALB ($16/month)
├── Rule 1: Host=www.example.com → prod TG
├── Rule 2: Host=dev.example.com + SourceIP → dev TG
└── Default: 403 (blocks direct ALB access)

Savings: $16/month (eliminated second ALB entirely)

---

Cost Focus 10: Where teams usually get this wrong for secure delivery (Aws Cost Optimization)

Problem

Dev environment running 24/7 but only used during work hours (~8h/day, 5 days/week).

Solution

Lambda-based auto-shutdown after 30 minutes of zero traffic:

# Lambda function (triggered by CloudWatch Alarm via SNS)
import boto3

def handler(event, context):
ecs = boto3.client('ecs')
ecs.update_service(
cluster='my-cluster',
service='my-dev-service',
desiredCount=0
)
return {'statusCode': 200, 'body': 'Dev service scaled to 0'}

# CloudWatch Alarm
DevIdleAlarm:
Type: AWS::CloudWatch::Alarm
Properties:
MetricName: RequestCount
Namespace: AWS/ApplicationELB
Statistic: Sum
Period: 300
EvaluationPeriods: 6 # 30 minutes
Threshold: 0
ComparisonOperator: LessThanOrEqualToThreshold
TreatMissingData: notBreaching
AlarmActions:
- !Ref DevIdleShutdownTopic

Savings: ~$18/month (dev only runs during active use)

---

Cost Focus 11: The practical decision path for predictable operations (Aws Cost Optimization)

Discovery

aws ec2 describe-addresses --query "Addresses[?AssociationId==null]"

Found 2 unattached Elastic IPs from deleted resources. At $3.65/month each:

aws ec2 release-address --allocation-id eipalloc-EXAMPLE1
aws ec2 release-address --allocation-id eipalloc-EXAMPLE2

Savings: $7.30/month

---

Cost Focus 12: How to execute without guesswork for exam and field confidence (Aws Cost Optimization)

Untagged images accumulating in ECR:

# Set lifecycle policy to keep only last 5 images
aws ecr put-lifecycle-policy `
--repository-name my-app-frontend `
--lifecycle-policy-text '{
"rules": [{
"rulePriority": 1,
"description": "Keep last 5 images",
"selection": {
"tagStatus": "any",
"countType": "imageCountMoreThan",
"countNumber": 5
},
"action": {"type": "expire"}
}]
}'

---

Cost Focus 13: What to validate before shipping for cleaner ownership (Aws Cost Optimization)

Analysis

NAT Gateway was provisioned for private subnet egress but:

• Backend containers only need to reach RDS (within VPC)
• Frontend SSR doesn't make external API calls
• ECR pulls can use VPC endpoints

Solution

• Removed NAT Gateway
• Added VPC endpoints for ECR, S3, and CloudWatch Logs
• Moved ECS tasks to public subnets with assignPublicIp: ENABLED

Savings: $32/month (single largest win)

---

Reference checks for Aws Cost Optimization #73

Primary references used for verification:

• https://docs.aws.amazon.com/
• https://docs.github.com/