Choosing compute
Instance families are tuned for general purpose, compute, memory, storage or GPU workloads. Right-sizing from real utilisation data is the single biggest cost lever most teams never pull.
Purchase options
On-demand is flexible and expensive. Reserved instances and savings plans trade a one- to three-year commitment for large discounts. Spot/preemptible capacity is up to 90% cheaper but can be reclaimed with a short warning, so it suits stateless, checkpointed or batch work.
Scaling and availability
Horizontal scaling adds instances behind a load balancer and is the cloud-native default; vertical scaling grows one machine and eventually hits a ceiling. Spread instances across availability zones, and use regions for disaster recovery and data residency.
- Scale on a leading signal (queue depth, RPS), not just CPU.
- Set sane cooldowns to avoid thrashing.
- Health checks must fail fast and honestly.