wk at gnupg.org
Thu Oct 19 18:32:04 CEST 2000
I got a new hard disk and a Linux-2.4.0test1 kernel on a K6-450, 64k
cache, 128MB. I did some tests with large files and also got some
Encrypt 12m7 -
Decrypt 20m55 22m24
This has been done without compression, with MDC on 2.5G random data
(25 blocks of 100MB /dev/urandom stuff). Values are only user time.
We can see that Rijndael is somewhat faster than Twofish even that
the Rijndael implementaion has not yet been optimized. I lost the
Twofish encryption timime, but it should be nearly the same as the
one for Rijndael. Given that we use CFB mode and therefore only
using the cipher in encryption mode the different values are not
subject to the algorithm but due to other implementation issues like
blocking size and so.
I did another test decrypting Rijndael and I saved 7 seconds when
using --no-armor which does mean that the data is not passed through
the dearmor code stub. So the overhead does not seem to be the
internal filter functions but maybe in the CFB code which get's
badly aligned data. Just a guess.
Werner Koch GnuPG key: 621CC013
OpenIT GmbH http://www.OpenIT.de
More information about the Gnupg-devel