[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20060502180715.GA9561@openwall.com>
Date: Tue, 2 May 2006 22:07:15 +0400
From: Solar Designer <solar@...nwall.com>
To: john-users@...ts.openwall.com
Subject: Re: Performance comparison (was:Error while installing Jumbo Patch)
On Tue, May 02, 2006 at 07:27:17PM +0200, Frank Dittrich wrote:
> I just wanted to know what Pentium processor your system has
> (/proc/cpuinfo), and how much performance is gained by using
> an mmx-optimized version, compared to the non-mmx version from
I'll tell you: the use of MMX, compared to plain x86, results in 3-4
times better performance for DES-based crypt(3) on almost all CPUs.
The performance improvement for LM hashes is comparable, but it differs
from system to system.
The effect the use of MMX has on performance at Kerberos AFS hashes is
small (30% improvement for "Short" passwords and usually 0% to 50% for
"Long" ones). But this implementation is far from optimal.
Other hash types are unaffected (because currently there are no MMX
implementations of those).
> Since the compiler version (and perhaps even the Linux kernel
> version - at least the distinction 2.4 vs. 2.6) has an impact
> on performance, I asked for this information as well.
No, the Linux kernel version (and even whether this is Linux at all) is
almost irrelevant. With x86 builds, the gcc version is also almost
irrelevant for most of the officially supported hash types - because
most of the processor time is spent in code written in assembly. LM
hashes are the only exception here - the key setup is in C, and with
non-MMX builds this is in fact pure C.
> Others also frequently asked for a performance overview
> comparing different john versions, hardware platforms, ...
Yes. I might setup a web page with John benchmarks eventually.
> Because the output is from a VMware guest system, there might
> be some differences compared to directly running john on a Linux
> system, but the differences shouldn't be too large.
Actually, this time the differences are large. Apparently, there was
other load on the host system when the non-MMX benchmark was performed,
but this is not reflected in real vs. virtual c/s rate differences in
the benchmark output specifically because of the use of VMware.
--
Alexander Peslyak <solar at openwall.com>
GPG key ID: B35D3598 fp: 6429 0D7E F130 C13E C929 6447 73C3 A290 B35D 3598
http://www.openwall.com - bringing security into open computing environments
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.
