Should I use decimal (GB) or binary (GiB) units in estimates?

Decimal, always, in interviews and back-of-envelope work. 1 GB as 10^9 bytes keeps every conversion a matter of moving the decimal point. The 7 percent gap between GB and GiB is far smaller than the error bars on your assumptions, so binary units add friction without adding accuracy.

How accurate does a storage or throughput estimate need to be?

Within a factor of 2 or 3 is fine, because the decisions are order-of-magnitude decisions. Whether the answer is 40 TB or 90 TB, the architecture is the same; whether it's 400 GB or 400 TB, it isn't. Spend your precision on the assumptions that swing the total, usually record size and the ratio of media to metadata.

When do I actually need to shard a database?

When sustained write throughput or total data size exceeds what one node plus the standard mitigations can hold. Read pressure is solved by caches and replicas without sharding. If writes stay under a few thousand per second and data under a few TB, sharding buys you operational pain and no capacity you needed.

Numbers to Knowchapter 2 of 4 · 3 lessons

Storage & Throughput

Reference numbers for back-of-envelope storage and bandwidth calculations.

← All chapters View in cheatsheet

Your product stores messages, and somebody asks how much disk you need for the year.

You do the sum in your head. A message is about 200 bytes, you handle 2 million a day, so 400 megabytes a day and about 150 gigabytes a year. Buy a terabyte and stop worrying.

Twelve months later the disk is full. The real figure was a little over 2 terabytes.

Your 200 bytes was the text of the message. It did not count the sender, the recipient, the timestamp, the read receipt, the two indexes over that table, the replica holding a second copy, or thirty days of backups. Every one of those is real disk, and together they were most of it.

Lessons

3 in this chapter

Sizing Data HonestlyDerive object sizes from their contents, then multiply by the overhead everyone forgets.3 min
What One Machine Can DoSingle-node throughput ceilings tell you when sharding is real and when it's resume-driven.3 min
From Estimates to ArchitectureThe point of the arithmetic is a decision: one box or many, cache or not, CDN or origin.3 min