Re: symbol catenation and montgomery

[Werner Koch]

On Fri, 29 Sep 2000, Torbjorn Granlund wrote:

> It is frustrating to me if GPG's speed gives GMP poor reputation.  I
> think GPG uses a modified version of GMP 2, which could explain part
> of its performance problems.

It is based on GMP 2.0 and has some modifications:

 * Only the mpn stuff is used, this is to save space and reduce code
   complexity.
   
 * The mpi objects are just a pointer to so some opaque data, this
   is for ease of implementation, don't have the MPIs on the stack,
   makes some operations much easier.  This is not a performance
   issue becuase GnuPG actually does only do very few calculations
   and most time is spend inside the math functions.
   
 * The library has further been changed to distinguish between
   confidential values (which are stored in another memory area)
   and normal numbers.  My estimation is a about 10% performance
   penalty which doesn't matter.  (We are not doing any cracking
   contest)
   
I'd like to hear about any performance problems.  I have compared
GnuPG to PGP 6.5 and they perform at about the same speed related to
cryptographic operations.  Access to large keyrings is indeed slow and
we are addressing this in 1.1.

Ciao,

   Werner
   
-- 
Werner Koch                             GnuPG key:  621CC013
OpenIT GmbH                             http://www.OpenIT.de