Re: [Asrg] 6. Email Path Verification (hashcash benchmarks)

> > > Please run the C hashcash code found at hashcash.org.  I would 
> > > appreciate seeing
> > > the performance numbers.  I would also like to see them for any non 
> > > Intel
> > > machines as well like Sparc and Alpha.
> > I'll see what I can pull together.
> First, a 1600MHz Athlon-XP:
>
> chromi(_at_)hydrogen hashcash-0.27 $ ./hashcash -s -b 20
> time estimate: 2 seconds
> chromi(_at_)hydrogen hashcash-0.27 $ ./hashcash -s
> speed: 649351 collision tests per second
>
> Second, a 200MHz Pentium-MMX:
>
> chromi(_at_)lithium hashcash-0.27 $ ./hashcash -s -b 20
> time estimate: 15 seconds
> chromi(_at_)lithium hashcash-0.27 $ ./hashcash -s
> speed: 69396 collision tests per second
>
> Both of the above are using GCC3 with default compile options, no 
> architecture or CPU specific stuff.  Surprisingly, the scaling looks 
> linear with clock speed, an old Pentium isn't usually this good when 
> compared with an Athlon.  I'll have to examine the algorithm and 
> figure out why.
> It doesn't compile first time on the Macs, will need to hack on it a 
> bit.
Got it to compile on a Classic Mac, mostly by hacking out all the 
option parsing.  The below is using CodeWarrior Pro 5, optimised for a 
PPC601 or 68020.
400MHz G3:    270000 collisions/sec
400MHz G3:    114000 when emulating 68K code
 80MHz 601:    36100
 80MHz 601:    13700 when emulating 68K code
 80MHz 68040:   5500
 16MHz 68030:   1000

Plus, I had to un-stiction the HDs on the two 68K machines.  They'd 
been sitting around for a while.
So, broadly speaking, the IIcx will take about 20 minutes to make a 
20-bit hashcash, on average.  That's a lot better than I expected, and 
it *might* just be acceptable to patient old Granny, as long as it 
doesn't happen while she's already online, and she has the presence of 
mind to go make a pot of tea or something while she waits.
It looks like the '030 isn't as far behind the '040 (architecturally) 
as I thought, though it still has a big clock-speed gap.  I also take 
these numbers as a fairly clear sign that the hashcash algorithm fits 
into a very small L1 cache, because the IIcx's memory is *very* slow.
(NB: Apple marketed the 80MHz 68040 as a "40MHz" CPU, referring to the 
external clock rather than the clock-doubled core.  Modern terminology 
always refers to the core speed, so I corrected that here.  The '040 
can be considered equivalent to a 486DX2.)
Finally, the SPARC machine - which I'm still unsure of the specs for - 
gave about 250000 collisions/sec, and an estimated 4 seconds to make a 
20-bit hashcash.  If anyone who knows about Solaris can tell me how to 
extract basic information about the machine (from an unprivileged 
account, natch), I'd be grateful.  I'd hazard a guess that it's a 
recent but low-end multiprocessor workstation.
I'd also try to give you scores for an ARM, but I don't have access to 
my parents' Acorn Archimedes from over here.
--------------------------------------------------------------
from:     Jonathan "Chromatix" Morton
mail:     chromi(_at_)chromatix(_dot_)demon(_dot_)co(_dot_)uk
website:  http://www.chromatix.uklinux.net/
tagline:  The key to knowledge is not to rely on people to teach you it.


_______________________________________________
Asrg mailing list
Asrg(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/asrg