[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Tue, 8 Jul 2008 03:26:33 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: wiki updates - benchmarks
On Mon, Jul 07, 2008 at 02:44:32PM -0600, RB wrote:
> > BTW, another recently updated page is the one with benchmarks:
> >
> > http://openwall.info/wiki/john/benchmarks
>
> Looks great! Do I presume correctly that you would prefer only
> significantly different results, or is a simple "be reasonable" in
> order? For example, a quick and dirty test shows gcc-4.1.2 on Linux
> doesn't provide what I would think are notably different results on a
> P-III 1 GHz. Would you rather add those in due to the major compiler
> revision difference (and potentially forestall questions) or just
> leave them off since they're not much different?
This is a good question, and a non-trivial one for me. After some
thought, I've decided that it makes sense to submit/publish results that
either are or could potentially be significantly different. In your
example, the results could not potentially be significantly different
(well, unless things would have gone awry), because you were
benchmarking almost exclusively assembly code, not C. So instead of
having lots of almost identical benchmarks on x86 (for the same x86 CPUs
and clock rates), differing only in C compiler choice and/or version,
I've added a note explaining that these are not expected to differ much
(and why not).
In fact, I added an entire section - "What to (not) submit" - which
addresses your question above and some others ("unstable" performance
of LanMan hashes, indirect effect of operating system choice).
Wanted: benchmarks on Core 2 beyond 2.4 GHz (both "official" and
overclocked, with any overclocking indicated as such). I know that some
people are running these CPUs at clock rates as high as 4+ GHz.
Benchmarks on more non-x86 CPUs, especially on fast ones (such as
PowerPC G5 at 2.7 GHz in some PowerMacs, and even newer POWER processors
at even higher clock rates in IBM servers). (I have many slow non-x86s
right here, but have no time to boot those systems up right now.)
Benchmarks on historical systems - Pentium MMX, the original Pentium,
486, and down to 386 at 8 MHz; then VAX and PDP-11. (No time to boot
one of my MicroVAXen up right now just to add a line to the table; maybe
someone else will do that.)
Benchmarks on unusual systems, such as Apple iPhone. (Yes, folks are
running JtR on their iPhones. It's terribly slow - but maybe not if you
compare to some historical systems.)
Unfortunately, JtR requires that "ints" be at least 32 bits wide, so no
benchmarks on 16-bit and 8-bit systems are possible unless you find a
weird compiler and/or tweak the code...
Thanks,
Alexander
--
To unsubscribe, e-mail john-users-unsubscribe@...ts.openwall.com and reply
to the automated confirmation request that will be sent to you.
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
