Cost-Optimizing AWS Infrastructure: A Case Study

AWS costs can quickly spiral out of control as applications scale. In this case study, we'll share how we reduced a client's monthly AWS bill from $28,000 to $16,000—a 42% reduction—without sacrificing performance or reliability.

Initial Assessment

The client was a B2B SaaS company running a monolithic application on AWS. Their infrastructure included:

12 EC2 instances (mostly m5.xlarge) running 24/7

RDS PostgreSQL db.r5.2xlarge with read replicas

Elastic Load Balancer and CloudFront CDN

S3 storage for user uploads (largely unoptimized)

CloudWatch and various other AWS services

The monthly bill breakdown:

EC2: $9,200

RDS: $6,800

Data Transfer: $4,200

S3 Storage: $3,100

Other services: $4,700

Optimization Strategy

We implemented a multi-faceted approach targeting the biggest cost drivers.

1. Right-Sizing EC2 Instances

**Finding:** Most EC2 instances were over-provisioned. CPU utilization averaged 18-25%, and memory utilization was below 40%.

**Action:** We analyzed CloudWatch metrics over 30 days and right-sized instances:

Downgraded 8 instances from m5.xlarge to m5.large

Converted 4 instances to t3.medium with burstable performance

Saved: $3,200/month

2. Spot Instances for Non-Critical Workloads

**Finding:** Development, staging, and batch processing workloads didn't require 24/7 availability.

Action:

Moved dev/staging to Spot instances (70-90% savings vs on-demand)

Configured auto-scaling for graceful handling of interruptions

Implemented scheduled shutdown for non-business hours

Saved: $2,100/month

3. Database Optimization

**Finding:** The RDS instance was oversized, and read replicas were underutilized.

Action:

Optimized slow queries (average query time reduced 65%)

Downgraded from db.r5.2xlarge to db.r5.xlarge

Removed one read replica (consolidated read traffic)

Implemented aggressive query caching with Redis

Saved: $2,800/month

4. Serverless Migration

**Finding:** Several API endpoints had sporadic traffic patterns but ran on dedicated EC2 instances.

Action:

Migrated 12 low-traffic API endpoints to Lambda

Moved static asset serving to S3 + CloudFront

Implemented Lambda@Edge for some dynamic content

Saved: $1,600/month

5. S3 Storage Optimization

**Finding:** 73% of S3 data was rarely accessed, and many objects were stored in Standard tier unnecessarily.

Action:

Implemented S3 Intelligent-Tiering for automatic cost optimization

Created lifecycle policies to move old data to Glacier

Enabled S3 compression for text-based files

Cleaned up incomplete multipart uploads

Saved: $1,400/month

6. Data Transfer Optimization

**Finding:** Significant data transfer costs between regions and to the internet.

Action:

Consolidated resources in a single region where possible

Implemented CloudFront caching more aggressively (cache hit rate improved from 62% to 89%)

Enabled compression for API responses

Saved: $1,800/month

Results Summary

Total monthly savings: **$12,000 (42% reduction)**

| Category | Before | After | Savings |

|----------|--------|-------|---------|

| EC2 | $9,200 | $5,000 | $4,200 |

| RDS | $6,800 | $4,000 | $2,800 |

| Data Transfer | $4,200 | $2,400 | $1,800 |

| S3 Storage | $3,100 | $1,700 | $1,400 |

| Other | $4,700 | $3,700 | $1,000 |

| **Total** | **$28,000** | **$16,800** | **$11,200** |

Performance Impact

Despite the aggressive cost cutting, we maintained or improved all key performance metrics:

**API Response Time:** Improved by 12% (thanks to query optimization)

**Uptime:** Remained at 99.9%

**Error Rate:** Unchanged at <0.1%

**Page Load Time:** Improved by 8% (better CloudFront caching)

Lessons Learned

1. **Monitor Everything:** CloudWatch metrics were crucial for identifying over-provisioned resources

2. **Start with Quick Wins:** Right-sizing and S3 optimization provided immediate ROI

3. **Test in Staging:** We tested all changes in staging before production

4. **Automate Where Possible:** Infrastructure as Code (Terraform) made changes repeatable and safe

5. **Continuous Optimization:** We implemented monthly cost reviews to prevent future bloat

Ongoing Optimization

We continue to optimize by:

Reviewing AWS Trusted Advisor recommendations monthly

Using AWS Cost Explorer for spend analysis

Implementing more aggressive auto-scaling policies

Evaluating new AWS services (e.g., Graviton instances for ARM workloads)

Conclusion

A 42% cost reduction is possible without compromising performance. The key is systematic analysis, incremental changes, and continuous monitoring. For most AWS customers, there are low-hanging fruit that can deliver significant savings with minimal risk.

If your AWS bill is growing faster than your revenue, it's time for an infrastructure audit.