[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Auth algorithm measurement tools?
> From owner-ipsec@portal.ex.tis.com Tue Jul 22 16:46:02 1997
> From: Dan.McDonald@eng.sun.com (Dan McDonald)
> Subject: Auth algorithm measurement tools?
> To: ipsec@tis.com
> Date: Tue, 22 Jul 1997 16:41:39 -0700 (PDT)
> Cc: ho@earth.hpc.org
>
> Hello!
>
> Apart from the obvious:
>
> get_time()
>
> /* Run test */
>
> get_time_again()
>
> print_time_difference()
>
> are there any good test harnesses for measuring algorithms like SHA-1 and MD5
> out there?
For measuring stand-alone performance, there's the
MD5 optimized package, including a modified wrapper, in:
ftp://ftp.isi.edu/pub/hpcc-papers/touch/md5-opt.tar.Z
see also
http://www.isi.edu/atomic2/md5-opts.html
This includes:
optimized source code (assembler is faster)
unrolled code
state variables forced to revisters
avoid copies for little-endians
'optimized' byte-reordering source code
the test 'wrapper' from the MD5 RFC, modified to:
use getrusage(), rather than time()
command-line variables for
block size
block repeat
double-buffer blocks (to force cache misses)
random block initialization (for data dependent algs)
skip block initialization
print bits/sec
Note - some of the fastest measurements have been for in-cache blocks
(on-chip cache), and drops of around 25% have been noted when data
is not in-cache.
Also, these kinds of 'wrapper' tests are not accurate predictors of
actual in-situ performance in IPv4. Tests in IPv4 indicate that
the packet processing can consume 2/3 of the processor capacity,
i.e., resulting in 1/3 the 'stand-alone' throughput for some algorithms.
FYI.
Joe
----------------------------------------------------------------------
Joe Touch - touch@isi.edu http://www.isi.edu/~touch/
ISI / Project Leader, ATOMIC-2, LSAM http://www.isi.edu/atomic2/
USC / Research Assistant Prof. http://www.isi.edu/lsam/