Non-cryptographic hashes like MurmurHash3 and xxHash are almost exclusively designed for hash tables, but they seem to work comparable (and even more profitable) to CRC-32 , Adler-32, and Fletcher-32 . Non-crystalline hashes are often faster than CRC-32, and produce a more "random" output, similar to slow cryptographic hashes (MD5, SHA). Despite this, I always see that CRC-32 or MD5 is recommended for data integrity / checksum.
In the table below, I tested 32-bit checksums / CRC / hash functions to determine how well they detect small differences in data:
The results in each cell mean: A) the number of collisions found and B) the minimum and maximum probability that any of the 32 output bits will be 1. To pass the test B, max and min should be as close as possible to 50. All that less than 45 or more than 55, indicates an offset.
Looking at the table, MurmurHash3 and Jenkins lookup2 are compared with CRC-32 (which actually does not give a single test). They are also well distributed. DJB2 and FNV1a are tested for collision, but not very well distributed. Fletcher32 and Adler32 work with NullBytes and 8RandBytes tests.
, : , " " ? - CRC-32/Adler-32/CRC-64 32-/64- ?