Cloud infrastructure cheatsheet
A one-page reference for cloud infrastructure design. For the full 3-tier HA reference architecture, see the complete guide.
📖 Full guide: Cloud Infrastructure →AZs vs regions
- AZ — isolated power/cooling/network within a region.
- Region — geographically separate, own control plane/compliance boundary.
Default to multi-AZ within a region; only go multi-region for a real RTO/RPO or data-residency requirement — it roughly doubles complexity/cost.
Active-active vs active-passive
| Active-active | Active-passive | |
|---|---|---|
| RTO | near-zero | nonzero (DNS/promote/warm cache) |
| Complexity | high (conflict resolution) | lower |
| Consistency | harder | easier |
Compute choice
| VMs | Containers | Serverless | |
|---|---|---|---|
| Cold start | slow | fast | varies |
| Ops overhead | high | medium | lowest |
| Best for | legacy/stateful | most services | spiky/event-driven |
Load balancing & autoscaling
- L4 LB: fast, IP/port only. L7 LB: content-aware routing, TLS termination.
- Autoscaling triggers on a leading indicator (queue depth, request rate), not just CPU — CPU often lags the thing you actually care about.
Disaster recovery
RTO = how long until you're back up
RPO = how much data you can afford to lose
Pick a DR pattern (backup/restore, pilot light, warm standby, active-active) based on your actual RTO/RPO, not the fanciest option.
Cost optimization
- Right-size instances — most workloads are over-provisioned.
- Reserved/committed-use for steady baseline load, spot/preemptible for interruptible batch work.
- Autoscale down aggressively off-peak, not just up on-peak.