System design: the 8 steps

A design round is graded on how you drive it. Say each step out loud before doing it, and keep an eye on the time.

1. Requirements	Functional: what must it do? Non functional: scale, latency, consistency, availability. Ask, don't assume.
2. Numbers	Users, requests per second, read to write ratio, storage per year. Round hard; you only need the order of magnitude.
3. API	Three to five endpoints or messages. Names and fields make the scope concrete.
4. Data	Tables or keys, the primary key, the indexes your queries need. Where each piece of data lives.
5. High level	Draw the boxes: client, load balancer, services, database, cache, queue. Walk one request through it.
6. Deep dives	Pick the 2 or 3 hardest parts and go deep. Ask: "Which part would you like me to go into?"
7. Failures	What breaks at 10x? What if the cache, the queue or a region dies? Retries, idempotency, backpressure.
8. Trade-offs	Say what you chose, what you gave up, and what you'd change at the next order of magnitude.

Numbers worth knowing

1 day	about 100k seconds, so 1M a day is about 12 per second
Memory read	about 100 ns
Redis or Memcached call, same zone	about 0.5 to 1 ms
Indexed Postgres or MySQL read	about 1 to 5 ms
Cross region round trip	about 50 to 150 ms
One Postgres primary	thousands to tens of thousands of simple writes a second
One Redis instance	about 100k simple operations a second
Availability	99.9% is about 43 minutes down a month, 99.99% about 4