Subhash Bohra

Posted on Dec 31

I Was Overpaying for AWS EC2 (Here’s What I Learned)

#aws #serverless #devops #costoptimization

Ever felt that uncomfortable pause before opening your AWS bill?

That moment where you know something is off, but you still hope it’s just a rounding error.

I’ve been there.

For years, EC2 felt like the safest option. Familiar. Predictable.

But in 2025, while managing a few internal services, I realized something uncomfortable:

I wasn’t paying for performance.

I was paying for idle compute.

The Moment It Clicked

Above: Real-world AWS cost dashboard highlighting idle EC2 capacity and fixed infrastructure costs.

Our setup looked reasonable on paper:

A small API service
A scheduled batch job
A webhook listener

All running on:

EC2
Application Load Balancer
Auto Scaling Groups

Traffic was unpredictable — short bursts, long quiet periods.

But the bill?

Consistently loud.

The Hidden Cost of “Always On”

When I dug deeper, a pattern emerged:

Instances were idle over 70% of the time
ALB and NAT Gateway costs never stopped
Nights, weekends, and off-hours were pure waste

That’s when it hit me:

EC2 wasn’t expensive.

Idle EC2 was.

STAR Breakdown (Interview-Ready)

⭐ Situation

Three internal services were running on EC2 with Auto Scaling, despite highly bursty traffic patterns.

⭐ Task

Reduce AWS infrastructure cost without sacrificing reliability or performance.

⭐ Action

Identified the least complex service
Refactored it into AWS Lambda
Exposed it using API Gateway
Implemented proper monitoring and secrets management

⭐ Result

Significant monthly cost reduction
No server patching or scaling rules
Cold starts consistently under 200ms
Fewer operational alerts and cleaner observability

This one change justified the entire experiment.

The Migration (What I Actually Did)

Above: Transition from EC2 + ALB + ASG to Lambda + API Gateway for bursty workloads.

Step 1: Break the Service Apart

Each Lambda function had a single responsibility.

Step 2: API Gateway as the Front Door

Clear request/response contracts, proper routing, and throttling.

Step 3: Observability from Day One

CloudWatch logs, metrics, and alarms were mandatory.

Step 4: Secrets Stayed Out of Code

Used SSM Parameter Store / Secrets Manager.

Step 5: Cold Start Control

Provisioned Concurrency only where latency truly mattered.

Sample Lambda Function (Node.js)


js
export const handler = async (event) => {
  console.log("Incoming request:", JSON.stringify(event));

  return {
    statusCode: 200,
    body: JSON.stringify({
      message: "Hello from Lambda!",
      requestId: event.requestContext?.requestId
    })
  };
};

DEV Community