Hacker News new | past | comments | ask | show | jobs | submit login

The question I care is: Is the probability of naturally occurring a collision in MD5 hashing significantly more than any other 128 bit hashing algorithm?

The answer is: No.




Sure, but if you're willing to accept lots collisions on "bad" input there are faster hashes (Bernstein and Jenkins have nice fast non-cryptographic hashes, for instance.)

Non-cryptographic hash functions may be a bad idea: e.g. don't store URI parameters (?foo=bar&baz=bar) in such a hash table, or you'll be vulnerable to rather simple DoS (this was all over the internet a week or two ago.)


> Non-cryptographic hash functions may be a bad idea: e.g. don't store URI parameters (?foo=bar&baz=bar) in such a hash table, or you'll be vulnerable to rather simple DoS (this was all over the internet a week or two ago.)

I think universal hashing is the usual protection against that kind of attack, and I think universal hashing is not considered cryptographic:

http://www.cs.rice.edu/~scrosby/hash/CrosbyWallach_UsenixSec...


> there are faster hashes

Yes, but none as widespread. MD5 is available in every system, on every language.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: