[hashcash] Re: Low end benchmarks

Yes the hashcash -s is measuring the wrong thing.  I didn't have
anything convenient to call in the loop so it is calling the generic
code still.  (Need something with a smallish ideally parameterizable
amount of work eg 100 collision attemps or 256 or whatever.  I just
call that in a loop count how many I managed to do until the clock
rolls over and then do the calculation.  Fastest way to measure with a
given clock resolution.  Not hugely accurate I'm finding, but at least
fast.)

The hashcash minting actually uses the correct Jonathan's libfastmint
hashcash core as selected by the static predictor code
(hashcash_select_minter()).

John: for what you're doing you need to run hashcash_benchtest, the
hashcash -s code is busted in hashcash-1.03p4* (all of the 1.03
versions).

Adam

On Sat, Aug 28, 2004 at 11:52:34PM +0100, Jonathan Morton wrote:
> >When I do 'make 386', I'm getting;
> >
> >knoppix@ttyp0[hashcash-1.03]$ hashcash -s
> >collisions = 36800
> >speed: 368000 collision tests per second
> >
> >On 'make x86-mmx', I get; speed: 375000 collision tests per second
> >
> >'make generic' gives me the best performance at 463,000 tests per 
> >second.
> >
> >Am I missing something?
> 
> Are you doing "make clean" in between?  If not, almost nothing will get 
> recompiled with the new settings.  The x86-mmx target should get you 
> about 2.6 million or so.
> 
> OTOH, the present method of doing a "quick benchmark" may well be 
> suspect.  I was going to put in a proper one, but haven't got around to 
> it yet.

Other related posts: