Principles of Content Addressing

Hash collisions are dependent on the length of the hash and the quality of the algorithm. AFAIK there has never been a _single_ SHA-1 has collision, and it's an old algorithm that people don't recommend using anymore.

As mentioned in the article, 12 bytes is enough to make collisions very unlikely. Assuming a high quality hash algorithm, you can use this table to determine the necessary length for a given number of documents: https://en.wikipedia.org/wiki/Birthday_paradox#Probability_t...