[Openswan dev] ocf-bench bug & performance :)

[Harald Jenny]

On Mon, Dec 20, 2010 at 09:17:40PM +1000, David McCullough wrote:
> 
> Jivin Paul Wouters lays it down ...
> > On Mon, 20 Dec 2010, Harald Jenny wrote:
> > 
> > >> [  460.738971] OCF: 100035 requests of 1488 bytes in 49 jiffies (2430.238 Mbps)
> > >
> > > Did you use a HW driver or just cryptoapi?
> > 
> > Just cryptosoft. No HW driver. (no cryptoapi, but ocf)
> 
> Yeah,  but cryptosoft is calling into cryptoapi to do the actual crypto.

Good I already thought I was wrong ;-).

> 
> So basically this is an in kernel benchmark of the OCF API using only the
> in kernel cryptoapi to do software based encrption across all 8 cores.

Hmmm impressive numbers nontheless...

> 
> In theory this is how fast ipsec could go if klips added no overhead.
> I think this should be goood enough to comfortably get us to 1Gbps of
> ipsec throughput.  Only a real test will tell, SMP performance is fickle and
> any misbehaviour with klips could blow this out.  It will be good to see
> the real ipsec results, for better or worse :-)

I'm already very keen to see them.

> 
> Cheers,
> Davidm

Kind regards
Harald

> 
> -- 
> David McCullough,      david_mccullough at mcafee.com,  Ph:+61 734352815
> McAfee - SnapGear      http://www.mcafee.com         http://www.uCdot.org