Re: Extensible hashing and RCU

[Evgeniy Polyakov <johnpol@2ka.mipt.ru>]

On Tue, Feb 20, 2007 at 08:17:31PM +0100, Eric Dumazet (dada1@cosmosbay.com) wrote:
> I shown your test was bogus. All your claims are just bogus.
> I claim your 'true random data' is a joke. rand() in your program is a pure 
> joke.

Care to reread your mail about your true random case with hash chain
length of 3 and 4? Anyway, I just shown that jenkins hash is simple to
crack and to find its collisions - even if you will put there some
constant value it will be the same. It is math, not something special
speculation about input values.

> Given 48 bits of input, you *can* find a lot of addr/port to hit one 
> particular cache line if XOR function is used. With jhash, without knowing 
> the 32bits random secret, you *cant*.

You seems to do not want to understand that it is exactly the same as
searching for collision law. It is simple, and results will be
dangerous.

> Again, you dont take into account the chain length.
> 
> If all chains were of length <= 1, then yes, xor would be faster. In real 
> life, we *know* chain length can be larger, especially in DOS situations.

I.e. you propose to add a hash, which has broken case for the same ip
addresses and different ports compared to good xor?
It was shown that hash(const, const, non_const) ends up with _broken_
distribution comapred to xor hash.

-- 
	Evgeniy Polyakov