Question 1

When should I start worrying about database scaling costs?

Accepted Answer

Start planning when your database exceeds 10GB or 10M rows, or when query response time averages over 100ms. Early warning signs: 50%+ CPU utilization at off-peak hours, query times degrading over time, connection pool exhaustion. The time to address scaling is before you hit limits — at 80% capacity — not after queries start failing. Implement read replicas, indexing, and connection pooling before they become urgent.

Question 2

What is the cheapest database for scale?

Accepted Answer

DynamoDB has the lowest cost per QPS at scale (request-based pricing, no idle costs) but becomes expensive for scan-heavy workloads. PostgreSQL on Neon or PlanetScale (serverless) has near-zero idle costs and scales read replicas efficiently. MongoDB Atlas is typically 20–40% more expensive than self-managed Postgres for equivalent workloads. For cost optimization at scale: start with managed Postgres, add read replicas, implement Redis caching, then evaluate whether NoSQL migration makes sense.

Question 3

How does data growth affect database costs?

Accepted Answer

20% monthly growth = 8.9× data in 12 months (compounding). This means your database will need 8.9× more storage, and potentially more compute (more data = slower queries without proper indexing). Costs typically scale faster than linear because: (1) Storage costs grow linearly, (2) Compute costs grow super-linearly once you exhaust single-node capacity, (3) Backup costs grow proportionally. Archive cold data (over 6 months old) to object storage — typically 10–20× cheaper per GB.

Question 4

What database indexes do I need at scale?

Accepted Answer

Critical indexes for scale: (1) Primary key — already indexed automatically, (2) Foreign keys — not auto-indexed in many DBs, costs full table scans on joins, (3) Frequently queried columns (WHERE clauses), (4) Composite indexes for multi-column queries (order matters: most selective column first), (5) Covering indexes for queries that need only a few columns — prevents table lookups. Rule of thumb: any query taking over 100ms in production needs an index audit. Use EXPLAIN ANALYZE in Postgres to diagnose missing indexes.

Question 5

Should I use SQL or NoSQL for my application?

Accepted Answer

SQL (Postgres/MySQL) wins when: Data has clear relational structure, ACID transactions required, ad-hoc query flexibility needed, strong consistency critical. NoSQL wins when: Document-oriented data with flexible schema, extreme write throughput required (DynamoDB at millions of WPS), horizontal sharding across regions (MongoDB). The modern consensus: start with Postgres — it handles 95% of use cases, scales to hundreds of millions of rows on a single node, and has better tooling. Migrate to NoSQL only when Postgres becomes a genuine bottleneck.

Database Scaling Cost Calculator — Project DB Costs as You Scale

Frequently Asked Questions

From the Blog