Hi jamal,
On 05 Apr 2005 12:11:35 -0400, jamal <hadi@xxxxxxxxxx> wrote:
> Hi Wang,
>
> On Tue, 2005-04-05 at 10:18, Wang Jian wrote:
>
> > I am not saying that we must use jenkins. We may use a less expensive
> > hash function than jenkins, but better than & 0xFF.
> >
>
> Sure; point is as long as it doesnt destroy the common use in place.
>
> >Anyway, I have done userspace test for jhash. The following test
> > is done in a AMD Athlon 800MHz without other CPU load.
> >
>
> No, the test i was asking for is to show distribution of the
> hash not how long it took (which is also an ok test).
Sorry, the test is not on the AMD Athlon 800MHz, but a 2.4G Pentium4! I
didn't notice that. So the jhash's performance is not fast enough.
On AMD 800Mhz
line speed @ 1000Mbps
[root@home ~]# time ./a.out
0.31user 0.00system 0:00.32elapsed 97%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+83minor)pagefaults 0swaps
line speed @ 100Mbps
[root@home ~]# time ./a.out
0.03user 0.00system 0:00.03elapsed 97%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+83minor)pagefaults 0swaps
>
> i.e if you fed the jenkins hash with 256 buckets - lets pick the number 1024
> samples of the data you showed earlier for how fwmark looks like,
> how well would the result look like.
> And what if you fed it with something like 1024 incremental fwmark from
> say 1..1024?
>
The test result looks not good. See attached file.
So let's find a better way.
--
lark
jhash-dist.txt
Description: Binary data
|