Database Economics: Provisioned vs Serverless
Learn the principles of NoSQL Capacity Units, why a simple loop can cost your company $10,000 overnight on AWS DynamoDB, and exactly when to use a standard SQL cluster.
The Serverless Billing Trap
In a traditional relational Database (like a standard MySQL droplet), you pay $50 a month for the server itself. Whether you query it 1 time, or 10 million times, the $50 bill never changes (unless the server physically crashes from CPU exhaustion). But modern cloud platforms push you towards Serverless NoSQL (like DynamoDB or Firestore).
Serverless databases never crash. They automatically scale to infinity. However, you pay strictly per operation. If you accidentally write a bad `while()` loop that accidentally updates every row in your database millions of times, AWS will instantly bill you tens of thousands of dollars.
Standard AWS Serverless Math
- Storage Cap Ex: $0.25 cents per GB stored forever per month. Usually irrelevant.
- RRU (Read Units): $0.25 per 1,000,000 Read Operations. Note: An operation is billed per full 4KB. If your row is 5KB, a single read actually costs you 2 operations.
- WRU (Write Units): $1.25 per 1,000,000 Write Operations. Writes are brutally expensive and are billed per single 1KB. If you write an 8KB blob of JSON, you are billed 8 discrete Write operations.
When to transition OFF Serverless?
Serverless throughput is incredibly cheap for small startups. But there is a mathematical apex. Usually, if your sustained traffic exceeds 150+ Requests Per Second explicitly, the mathematical on-demand invoice (e.g., $1,000/mo) will exceed the cost of just manually spinning up a massive dedicated EC2 Instance Cluster ($400/mo) to run standard Postgres.