SQLite Forum

How much would you trust a page-level checksum ?
Login
>I believe a 64-bit checksum is too small.

> Birthday paradox says (roughly) that given random numbers in 0 ... N, you should expect duplicates to show up when you have roughly sqrt(N) numbers.

With my trusty calculator<sup>1</sup>, I worked out how often a random collection of bytes would have the same 64-bit checksum as another random collection. Within 10%, if today's population on Earth were to toss the figurative dice to get another random collection and checksum every second, then among the whole population we could expect to see a match about every 74 years.

I'm not seeing how the Birthday Paradox applies to such infinitesimal odds. But I am not going to be sneering at those who settle for 64-bit check words. Somewhere in the assessment of diminishing return on investment, one should account for meteor strikes or man-made big events.

----

1. 2^64 / (24\*60\*60 \* 365) / 7.9e9