Re: [Ocf-linux-users] why aes algorithm faster than null algorithm when running cryptotest using cryptosoft?

[David M. <Dav...@se...>]

Jivin da xiao lays it down ...
> when I test ocf,I found this strange problem.I use:
> linux-2.6.26.8
> crypto-tools-20080917
> ocf-linux-20080917
> openswan-2.6.16
> 
> #
> # OCF Configuration
> #
> CONFIG_OCF_OCF=y
> CONFIG_OCF_RANDOMHARVEST=y
> CONFIG_OCF_CRYPTODEV=y
> CONFIG_OCF_CRYPTOSOFT=y
> 
> to run the test,I get the result:
> ./cryptotest -a null 256 8192
>    0.067 sec,     512   null crypts,    8192 bytes, 63046643 byte/sec,
> 481.0 Mb/sec
> ./cryptotest -a aes 256 8192
>    0.036 sec,     512    aes crypts,    8192 bytes, 117038368 byte/sec,
> 892.9 Mb/sec
> ./cryptotest -a md5 256 8192
>    0.008 sec,     256    md5 crypts,    8192 bytes, 272215992 byte/sec,
> 2076.8 Mb/sec
> ./cryptotest -a des 256 8192
>    0.078 sec,     512    des crypts,    8192 bytes, 53531550 byte/sec,
> 408.4 Mb/sec
> ./cryptotest -a 3des 256 8192
>    0.225 sec,     512   3des crypts,    8192 bytes, 18631166 byte/sec,
> 142.1 Mb/sec
> ./cryptotest -a sha1 256 8192
>    0.023 sec,     256   sha1 crypts,    8192 bytes, 90045170 byte/sec,
> 687.0 Mb/sec
> 
> why aes and md5 are so faster than null algorithm?

Beats me :-)

The null implementation in the kernel uses memcpy(src,dst,len),  which "may"
be slower than an optimised alg.

Could be an accounting bug in cryptotest.

I see the same thing here on a mips board so at least it's consistent :-)

Other than that I have no ideas, I would have tried it with "openssl speed"
but it doesn't support NULL,

Cheers,
Davidm

-- 
David McCullough,  dav...@se...,  Ph:+61 734352815
McAfee - SnapGear  http://www.snapgear.com                http://www.uCdot.org