On Mon, Nov 29, 2010 at 11:12:21AM -0500, der Mouse wrote: > > If the keys are controlled by a third party, it is very easy to > > degrade the performance to a linear list. > > Sure, but that's not a useful remark. It's equally true of (n*K)>>32, > or for that matter any other easily invertible hash function. If the > bucket count is small enough to make guessing feasible for the > attacker, it's true of any hash function cheap enough to be useful as a > hash function.
Only if you keep K fixed. If you follow the "rehash with different K on high number of collissions" scheme, it is resilient to such attacks. Joerg
