Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 23 Nov 2015 17:48:33 +0000
From:      RW <rwmaillists@googlemail.com>
To:        freebsd-questions@freebsd.org
Subject:   Re: cryptodev HW (aesni) vs software
Message-ID:  <20151123174833.0e36619c@gumby.homeunix.com>
In-Reply-To: <20151123145326.63d3203b@nonamehost.local>
References:  <20151120200325.2baade9c@nonamehost.local> <20151120192920.119bbf91@gumby.homeunix.com> <20151122130329.6ea9b9c6@nonamehost.local> <86lh9qhtuw.fsf@WorkBox.Home> <20151123145326.63d3203b@nonamehost.local>

next in thread | previous in thread | raw e-mail | index | archive | help
On Mon, 23 Nov 2015 14:53:26 +0200
Ivan Klymenko wrote:

> On Sun, 22 Nov 2015 10:38:47 -0600
> Brandon J. Wandersee <brandon.wandersee@gmail.com> wrote:
> 
> > Ivan Klymenko writes:
> >   
> > > I hope the problem has now become more visible?    
> > 
> > No. State what you believe the "problem" to be. We need to know how
> > what actually happened is different from what you expected to
> > happen. 
> 
> Problem 1
> At two loaded modules cryptodev and aesni to encrypt selected slower

cryptodev is known to be slower than userland acceleration.


> Problem 2
> Without any loaded modules cryptodev and aesni encryption occurs at
> the same rate as in the loaded module aesni - which suggests that
> hardware encryption is absolutely not working.

If I'm understanding you correctly, it's also consistent with both cases
using userland acceleration.

The aesni kernel module provides AES-NI support for crypto/cryptodev in
the kernel, not in userland.

 
> Problem 3
> In the best case FreeBSD inferior Linux in encryption by as much as
> 23% at exactly the same hardware

I think hardware would be more than 23% faster than software.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20151123174833.0e36619c>