Database Instance Sizing
DB instances don't degrade gracefully — they cliff. Memory fills, IOPS saturate, connections queue. The MySQL `innodb_buffer_pool_size` that was fine at 10 GB craters at 50 GB. Model your workload — data volume, QPS, query complexity, connections — against engine-specific sizing formulas before the cliff finds you in production.
📐 Workload Profile
Select engine, data size, QPS, query complexity, connections. The engine computes vCPU, RAM, storage, and IOPS per workload profile. Cross-cloud instance recommendations with estimated monthly cost.
Database Instance Sizing: The Engineering Behind the Numbers
Selecting the right database instance size is one of the most consequential infrastructure decisions an engineering team makes. Under-provision and you get P99 latency spikes, connection exhaustion, and OOM kills. Over-provision and you waste 40–60% of your cloud bill on idle resources — the single largest line item in most SaaS infrastructure budgets.
How the Sizing Engine Works
This calculator applies engine-specific formulas derived from production benchmarking and vendor best practices:
| Engine | Memory Formula | CPU Formula | Key Reference |
|---|---|---|---|
| PostgreSQL | shared_buffers (25% RAM) + work_mem × max_connections + OS cache | 1 vCPU per ~1,000 QPS (moderate queries) | Percona PG Tuning Guide |
| MySQL / MariaDB | innodb_buffer_pool_size (70–80% RAM) + per-connection overhead (~2–4 MB/conn) | 1 vCPU per ~500–800 QPS (InnoDB) | MySQL Performance Schema |
| MongoDB | WiredTiger cache (50% of RAM − 1 GB) + working set must fit in RAM | 1 vCPU per ~2,000 ops/s (document CRUD) | MongoDB Production Notes |
| Redis / Valkey | Entire dataset + overhead (~2× raw data for internal structures) | Single-threaded per instance; scale via clustering | Redis Latency & Sizing Docs |
Why "Just Use RDS" Isn't Enough
Cloud provider instance families (AWS RDS, GCP Cloud SQL, Azure DB) use incompatible naming conventions — db.r6g.xlarge vs db-custom-4-16384 vs Standard_D4s_v5. An engineer who knows they need "8 vCPU and 32 GB RAM" still has to manually decode each vendor's catalog. This calculator bridges that gap with mapped recommendations.