|
Hello,
On Fri, 20 Jun 2003, Wensong Zhang wrote:
> I didn't throw away the original SHIFTXOR hash, but multiply it with
> random value, so that it can avoid the hash attack. Please take time to
> evaluate it.
done: http://www.ssi.bg/~ja/lvs/test.txt
New tools: http://www.ssi.bg/~ja/hashlvs-0.3.tgz
New methods: SX (Shift and XOR, it is LVS+random mul), KNV
Results:
- jhash performs well with -O3, same speed as "1" on my CPU
- KNV is slow
- SX is bad for rnd = 0 (bad luck, all in one list)
Note that in all my tests "rnd" is const, probably bad value.
While changing 'rnd' SX access time increases, especially for
the small hash sizes. Not every random value is good for
multiplication.
> The diff is for ipvs-1.1.6. There are three hash choices, users can choose
> as they want.
It seems all hash funcs work well, the difference is in the
speed. SX can have problems with big random values and 0.
> Regards,
>
> Wensong
Regards
--
Julian Anastasov <ja@xxxxxx>
|
