Why MySQL Single Tables Should Stay Below 20 Million Rows: Theory & Practice

01 Auto‑increment Primary Key Perspective

When the primary key is an int, its maximum value is 2³²‑1 (about 2.1 billion). If the primary key is a bigint, the theoretical limit is 2⁶⁴‑1, which is far beyond what disk can hold. A tinyint caps at 255, so inserting an ID of 256 would fail.

02 Data Page Perspective

In InnoDB, a table is stored in an .ibd file that is divided into 16 KB data pages. Each page contains a page number, forward/backward pointers, a checksum, a page directory, and the actual row data.

The pages are linked together using a B+‑tree. Leaf nodes store the actual rows, while non‑leaf nodes store index entries. Each level of the B+‑tree corresponds to one disk I/O.

To estimate capacity, assume a data page can store roughly 15 KB of row data after accounting for metadata. With a bigint primary key (8 bytes) and a 4‑byte page number, each page can hold about 1 280 pointers (X = 15 KB / (8 + 4)). If each row occupies 1 KB, a page can store 15 rows (Y = 15).

- For a B+‑tree of height N, the number of leaf nodes is X^(N‑1).
- Total rows M = X^(N‑1) * Y.

Applying the formula:

- Height N=2 (two‑level tree): M ≈ 1 280^(1) * 15 ≈ 19 200 rows.
- Height N=3: M ≈ 1 280^(2) * 15 ≈ 2.5 million rows.
- Height N=4: M ≈ 1 280^(3) * 15 ≈ 3 billion rows.

In practice, with typical row sizes around 1 KB, a safe guideline is to keep a single MySQL table under 20 million rows to avoid performance degradation; this figure is a recommendation, not a strict rule.

03 Thought Exercise

Question: How many rows can a four‑level B+‑tree store if the primary key is bigint and each row is 1 KB?

Answer: Using X = 1 280 and Y = 15, the total rows M = 1 280^(4‑1) * 15 = 1 280³ * 15, which yields roughly 3 billion rows.

— End of article —