Re: [PATCH] nf_conntrack: optimize hash calculation of tuple

[Patrick McHardy <kaber@trash.net>]

Yasuyuki KOZAKAI wrote:
>>But anyway, this is just a possible micro-optimization, the
>>real question is why we need three invocations of jhash_mix.
> 
> 
> That patch is just the result of extracting jhash() and jhash_2words().
> I'm not familiar with hash algorithms. Does the degree of distribution
> become bad if we replaces 3 jhash_mix with something like XOR ?


I'd say no. jhash was chosen because it has good distribution even
for non uniformly distributed input values, so it shouldn't be
necessary to try to provide more uniformly distributed input values
by hashing multiple times.