ElGamal: key length vs performance

[Ryan Malayter]

On 4/2/06, Qed <qed at tiscali.it> wrote:
> Different implementations => different speeds.
> You cannot rely on a particular piece software to infer general
> performance figures for crypto algos.

This is very true. In my tests, for example, AES implementation in
GnuPG runs far slower than the implementation used in TrueCrypt, 7zip
or a number of other x86-specific programs.

I mentioned this speed difference to Werner a while back, and he
explained GnuPG has to work on many platforms, so using code optimized
for x86 - even if it is C-code optimized for x86 - isn't going to
happen. Which makes sense.

The easiest way to test is to simply encrypt the same file several
times using different --cipher-algo parameters on the command line. My
tests on Pentium 4s showed CAST5 to be the fastest algorithm in GnuPG
on that platform, but your own hardware is different, you should run
your own tests.
See this discussion at:
http://lists.gnupg.org/pipermail/gnupg-users/2005-August/026315.html