Capacity Planning

Auto-Scaling, Elasticity & Capacity Planning

Auto-scaling lets your fleet grow when traffic surges and shrink when it ebbs, so you pay for the load you actually have. This lesson covers reactive metric-based scaling, predictive (schedule-based) scaling, and the gotchas that turn auto-scaling into auto-outage: warm-up time, scale-down storms, downstream throttling, and cost runaway. We also walk through capacity planning: how to estimate the fleet size you need from QPS, latency targets, and headroom, before relying on the scaler to fix mistakes at 3 a.m. By the end you can configure an auto-scaling policy with confidence and explain to an interviewer why simply 'putting it on auto-scale' is not the actual answer.

Back-of-the-Envelope Estimation & Capacity Planning

Back-of-the-envelope estimation is the math you do in three minutes to ground a system design in numbers. It is what tells you whether your single Postgres instance can handle the load (no), how much storage you need over five years (probably more than you think), and how much CDN bandwidth you are about to commit to (probably more than that). This lesson covers the standard latency / throughput / size / bandwidth numbers every engineer should have memorized, the unit conversions and order-of-magnitude reasoning that keep you fast, the templates for QPS, storage, and bandwidth estimation, capacity planning beyond steady state (peak vs average, headroom, growth, regional, seasonal), and the cost rough-arithmetic that turns 'we need more servers' into a defensible business case. The goal is to leave you able to walk into any interview or design review and produce useful numbers in three minutes flat.

Behavioral for Backend / Infra Engineers

Backend and infrastructure engineering loops grade for a cluster of behavioral signals that frontend and product engineering loops weight less heavily: reliability and oncall judgement, capacity and scale thinking, data-integrity decisions under pressure, and the empathy-for-the-pager dimension that distinguishes engineers who can be trusted with production. The behavioral signal is most often woven into the system-design round and the oncall-and-incident round, with explicit story shapes (the 3am page, the SLO trade-off) that interviewers reach for. This lesson defines the cross-cutting backend signals interviewers grade, walks through how the loop folds the behavioral signal into the technical rounds, maps the signals to the questions interviewers ask, and shows two model answers tailored to the incident-response and capacity-planning story shapes.

Choose a hosted API or a self-hosted model

Compare a hosted model API with a self-hosted open-weight model for a live tutor and a nightly scoring job: apply a quality gate, compute monthly cost at real utilization and the breakeven volume, check a data-location rule, and say what would reverse the choice.