benchmarks

[Jussi Kivilinna]

On 09.12.2013 19:04, Werner Koch wrote:
> Hi,
> 
> I did some benchmark charts for Libgcrypt.  See
>   
>   ftp://ftp.g10code.com/g10code/scratch/gcrypt-bench-x220-2300.gnumeric
> 
> or just the charts:
> 
>   ftp://ftp.g10code.com/g10code/scratch/gcrypt-bench-x220-2300.pdf
> 
> this nicely shows Jussi's improvements.  The values for Nettle have been
> taken from the nettle-benchmark example program.  I did not add data for
> OpenSSL because it is a bit of work to extract the numbers from "openssl
> speed".  It would be nice to have wrapper for openssl so that
> bench-slope can be used to create compatible numbers.  

To compare with Nettle and OpenSSL, I have done some testing with a modified
version of crypto-speedtest from

 http://panthema.net/2008/0714-cryptography-speedtest-comparison/

Graph showing differences between OpenSSL (1.0.1e), libgcrypt-1.5,
libgcrypt-1.6 (ECB & CTR) and nettle (2.7.1) on Intel Haswell:

 http://koti.kapsi.fi/~jukivili/gcrypt/haswell-3200-ubuntu-saucy-gcrypt.pdf

Above graph shows additional benefit of new bulk processing implementations
for parallelizable modes of operation (CTR, CFB-dec, CBC-dec).

Each measurement are for key-setup(enc)+enc+key-setup(dec)+dec for different
buffer lengths. Default key-length is 256-bit; shorter used if 256-bits not
supported.

-Jussi

> 
> 
> Shalom-Salam,
> 
>    Werner
> 
>