cloud-computelisted

# Cloud Compute Compute is the engine of the cloud — the CPU, memory, and runtime that executes your workload. ## Instance sizing Right-size before you optimize. A `t3.medium` at 80% utilization costs less than a `m5.xlarge` at 15%. Measure actual usage before picking an instance family. Over-provisioning is the most expensive bug in cloud. ## Serverless vs. containers Lambda wins for bursty, event-driven workloads with idle periods. Containers win for steady-state, long-running services. The break-even is not CPU cycles — it is cold-start latency and the 15-minute timeout ceiling. If your function spends 3 seconds starting and 200ms running, you are paying for the wrong runtime. ## Auto-scaling Scale on a metric that leads demand, not one that lags it. CPU is lagging — by the time CPU hits 80%, requests are already queuing. Queue depth or request latency is leading. Set scale-in cooldowns longer than scale-out: tearing down is cheaper than spinning up, but flapping is worse than either. ## Spot and preemptible Use spot instances for stateless, fault-tolerant, interruptible workloads. Never run a stateful database on spot. The savings are real (60-90%) but the two-minute termination notice is real too — your application must handle it gracefully.