4 ms·

Hash functions: theoretical basics, mixing, and evaluations of algorithms

20 points by nreece 18y ago

michael_dorfman 18y ago
Excellent article, with a nice degree of depth. I'd like to see more by this author.
carterschonwald 18y ago
Umm, hash functions ARE NOT used for crypto, one way functions are, and they are not hash functions.
- sah 18y ago
  The phrase "hash functions" refers to a general category that includes "cryptographic hash functions" (which are hopefully one-way) as a special case.
ovi256 18y ago
Great page layout and graphic design! Peeking at the CSS, I notice that the main div is called "paper" and that the offset lengths are measured in centimeters and millimeters. I find that intriguing and unusual. Was the layout created with a WYSIWYG editor? Or is it due to the designer's preferences?
ardit33 18y ago
I think every programmer needs to know how Hash Function, and Hashpams/Tables work from the inside-out. It gives you a great insight on why certain techniques are used, and it is a classical CS problem. If you can write a simple but solid implementation of a hash function, and a hash map, you are ahead of most developers out there. A lot of developers use them everyday, but don't have the faintest idea on how they work.
ajross 18y ago
The author seems to have more depth in theory than in practice, unfortunately. The first function examined, and casually discarded (without attribution) is the Dan Bernstein djb2 function. This is a very popular function, largely because of its extreme simplicity ("multiply by 33 and add/xor in the next byte") and comparative robustness (it's VERY strong for a one-line algorithm). Neither of these benefits seem to be of interest, though, so he drops this one on the floor and he then trots off to examine a bunch of 20-30 line monsters. Which is interesting enough, but not terribly useful to someone who actually wants to implement, y'know, a real-world hash table. djb2 happens to be a favorite of mine, it's a great thing to keep in your head for when you need a nice robust mixing function. If you really have requirements that go beyond what it can provide, you're probably better off just using a cryptographic hash than trying to internalize the stuff in this article. Given that digest hashes are the strongest algorithms available anyway, all the advice here is just carefully disguised premature optimization, IMHO.

20 points by nreece 18y ago

michael_dorfman 18y ago

Excellent article, with a nice degree of depth. I'd like to see more by this author.

carterschonwald 18y ago

Umm, hash functions ARE NOT used for crypto, one way functions are, and they are not hash functions.

sah 18y ago
The phrase "hash functions" refers to a general category that includes "cryptographic hash functions" (which are hopefully one-way) as a special case.

ovi256 18y ago

Great page layout and graphic design! Peeking at the CSS, I notice that the main div is called "paper" and that the offset lengths are measured in centimeters and millimeters. I find that intriguing and unusual. Was the layout created with a WYSIWYG editor? Or is it due to the designer's preferences?

ardit33 18y ago

I think every programmer needs to know how Hash Function, and Hashpams/Tables work from the inside-out. It gives you a great insight on why certain techniques are used, and it is a classical CS problem. If you can write a simple but solid implementation of a hash function, and a hash map, you are ahead of most developers out there. A lot of developers use them everyday, but don't have the faintest idea on how they work.

ajross 18y ago

The author seems to have more depth in theory than in practice, unfortunately. The first function examined, and casually discarded (without attribution) is the Dan Bernstein djb2 function. This is a very popular function, largely because of its extreme simplicity ("multiply by 33 and add/xor in the next byte") and comparative robustness (it's VERY strong for a one-line algorithm). Neither of these benefits seem to be of interest, though, so he drops this one on the floor and he then trots off to examine a bunch of 20-30 line monsters. Which is interesting enough, but not terribly useful to someone who actually wants to implement, y'know, a real-world hash table. djb2 happens to be a favorite of mine, it's a great thing to keep in your head for when you need a nice robust mixing function. If you really have requirements that go beyond what it can provide, you're probably better off just using a cryptographic hash than trying to internalize the stuff in this article. Given that digest hashes are the strongest algorithms available anyway, all the advice here is just carefully disguised premature optimization, IMHO.