Estimate scale on the back of an envelope

Estimation turns “lots of users” into numbers that drive decisions. Does the data fit on one machine? Do reads need a cache? Is one database enough for the write rate? You are not expected to be precise - you are expected to be within an order of magnitude, quickly, and to use the result.

• A day has 86,400 seconds. For quick math, call it ~100,000 (10^5).
• Average QPS = daily requests ÷ 86,400. Peak QPS is often assumed to be 2-3× the average.
• Storage = writes per day × bytes per write × days retained.
• Sizes: 1 KB ≈ 10^3 bytes, 1 MB ≈ 10^6, 1 GB ≈ 10^9, 1 TB ≈ 10^12, 1 PB ≈ 10^15.
• Read:write ratio tells you which path to optimize. A social feed might be 100:1.

A worked example: 100 million daily users making 10 requests each is 1 billion requests a day. Exactly, that’s about 11,574 QPS on average and 23,148 QPS at a 2× peak.

With the 10^5 shortcut: 100 million users × 10 requests = 10^9 requests per day, divided by 10^5 seconds ≈ 10,000 QPS. That is close enough to decide that a single application server will not do, but a handful behind a load balancer will. Storage works the same way: 1 million posts a day × 1 KB × 365 days × 5 years ≈ 1.8 TB, which a single database can hold - so you may not need sharding for size alone.