Snowflake
- Auto-suspend warehouses after 60 seconds idle — often cuts compute spend 40–60%.
- Use result caching — identical queries within 24h cost zero compute.
- Query profile expensive queries. A 10-minute query is often fixable to 30 seconds with a filter push or materialised table.
S3
- Compress with Snappy or ZSTD — typical 5–8× size reduction from raw CSV.
- Right-size storage class with lifecycle policies — Standard-IA for data accessed less than monthly.
- Enforce column pruning — ensure query engines are not reading unused columns.
Spark
- Use Spot/Preemptible instances for batch jobs — 60–80% cheaper. Add checkpointing for resilience.
- Right-size executors — over-provisioned executor memory wastes money. Profile with Spark UI.
Kafka
- Set appropriate retention — not everything needs 7-day retention. Compacted topics for reference data.
- Enable compression (lz4 or snappy) at the producer level — reduces network and storage cost.