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

> >  Looks like I was wrong about it being low-end.  :-)  This machine
> >  replaced a cluster of three four-way 100MHz boxes, which *were*
> >  repurposed workstations.
> Four-way 100MHz boxes?  I don't recall Sun making something like that. 
>   What kinds of machines were they exactly?
I don't remember very precisely, but a friend who apparently saw them 
physically reported that they were of the "pizza box" form factor used 
for many Sun workstations.  A fourth, identical unit was used as a 
fileserver for the other three.
Needless to say, these machines weren't particularly fast, but they did 
handle lots of users quite well, except when a CPU conked out in the 
fileserver with fairly predictable results.  They can't have been 
enterprise servers, otherwise the admins would have been able to 
disable the bad CPU until the replacement arrived.
> >  FWIW, it appears that the UltraSPARC II is slightly less efficient
> >  (by maybe 20%) per-clock than the G3, but that's not a bad deal
> >  overall - the G3 is known to be exceptionally good in that area.
> Keep in mind that your E450 is an SMP box, but the code you're running 
> almost certainly is not parallelized.  So, you're only going to be 
> using one CPU.
Naturally.  I was comparing the 250K score on the 480MHz UltraSPARC 
with the 270K on the 400MHz G3, which works out about 20% difference.
But in a server environment, you're going to be able to search for 
several hashcashes in parallel, so the overall throughput of the E450 
will be about one hashcash per second, while the G3 will only be one 
every four.  The individual code doesn't need to be parallelised, you 
just need multiple jobs to run.
And it looks like my Pentium-MMX will take longer to make hashcash than 
it presently does to run SpamAssassin.  A data point worth noting.
> It would be interesting to compare these hashcash benchmarks against 
> the SPEC CPU benchmarks at <http://www.spec.org/cpu2000/>, as well as 
> the predecessors, so that we can try to get some sort of idea how 
> various classes of machines will perform.
Yes, that might be interesting.  Also worth checking are the 
distributed.net statistics, some of which are collected on a 
per-architecture basis.  The distributed.net stats are interesting 
mainly because they use a cryptographic algorithm (RC5), which is 
presumably not overly dissimilar in characteristics to hashcash (SHA1).
> If you can point me at the code, I can run the same test on a variety 
> of hardware I have access to, which may provide some useful additional 
> data points.
http://hashcash.org/

The actual code is on the "old site" linked from there.  As mentioned, 
I had to hack it up to make it work on Macs, but I think I have all the 
useful info we can get from Macs for now.
--------------------------------------------------------------
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