From: Scott A Crosby <scrosby(_at_)cs(_dot_)rice(_dot_)edu>
...
\epsilon. A hashcash scheme with an 'out' that allows random nonces
does not incur the full cost is still practical and requires little to
no CPU time.
I cannot see how any "out" that requires little or no CPU time for
unknown good guys could not be used by unknown bad guys. As I see it,
a bad guy can always pretend to be 1,000,000 strangers unless you
add some kind of authentication.
to buy equal speed CPUs? How else do you ensure that the spammers
don't have CPUs that are 10,000 times faster than spam targets?
This is unlikely. Current CPU's are 3ghz, who runs a 300khz CPU
anymore? Even a Commodore 64 or an Apple ][ is faster!
Do you think that the only influence on CPU speed is raw clock speed?
For example, a 25 MHz 386 has a raw clock speed only 100 times less
than a 2.5 GHz P4, but easily 1000 times slower. Look at the number
of those 40 usec cycles required by 386 CPUs for common instructions.
Then look at the effects of instruction caches.
I remember recently reading about a primitive, different from partial
hash collision, that is limited by memory bandwidth, not CPU
speed. Memory bandwidth varies much less that CPU performance.
Memory bandwidth varies by by even larger values if you look at real
systems instead of netnews cartoons. Main memory bandwidths of cheap
systems today only about 6 times faster than they were 10 or 15 years
ago, but memory size and caches more than make up the difference. On
the other hand, if your CPU burner has a big enough memory footprint
to bust GHz caches (including those with various burst fill schemes),
you'll need to worry about whether it can even run on old systems.
256 MBytes sounds small today, but inconceivably large for systems
of 15 years ago.
...
A competent programmer also tries to keep their numbers accurate
within two orders of magnitude. Fast CPU's are 3ghz. The origional XT
(way before my time) was 4.77mhz, or only 600x slower. Do there exist
many widely deployed systems email systems that are <25mhz? Thats only
120x slower than current machines, not the 10,000x as claimed.
That is evidence of other than competence, because raw clock speed is
merely one component of CPU speed. The original XT was a lot slower
than 600X slower than a 3 GHz P4 because it used an 8088 wiht an 8-bit
bus (good for 4X hit), many instructions took several cycles (good
for another ~4X hit) and there was no I or D cache (good for another
hit of 10-1000X on a pure CPU grinder). A 5 MHz 8088 based system
was slower than that value of 600*4*4*10 compared to a 3 GHz P4.
Moreover, my number of 10,000 include several future years of Moore's Law
cycles. I'm "only" claiming a current 1000X span of CPU speeds.
Vernon Schryver vjs(_at_)rhyolite(_dot_)com
_______________________________________________
Asrg mailing list
Asrg(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/asrg