Christian Hohnstaedt schrieb:
On Fri, Aug 22, 2008 at 04:27:02PM +0200, Tomasz Chmielewski wrote:
Christian Hohnstaedt schrieb:
The overhead of putting the data into the kernel and into the NPE
and back again was amortized by hardware speed only
above 1Kbyte of data length on the IXP4xx CPU.
In other words, it could be slower for some operations?
Yes, definitly for short data.
I was going to use it with IXP4xx CPU; and was wondering if it would be
of any help in OpenVPN traffic (which uses OpenSSL) on a loaded machine.
Whenever possible, use real VPN.
If you like proper VPN throughput on your IXP4xx box, use real VPN.
IPsec you mean? Sure.
There are some deployments where other types of VPN are more functional.
Which is off-topic here, anyway.
This was at least the result back then when I tested OCF-cryptodev
with my driver.
Do you have any benchmarks available?
For opensslwith ixp4xx hw crypto? No. But google has:
http://ocf-linux.sourceforge.net/benchmarks.html
Where you can see that the break-even is reached at ~256 byte packets.
(and not 1K as I remembered...)
Staying on theoretical grounds, it should be perfectly possible to pass
only packets bigger than X bytes to the kernel.
--
Tomasz Chmielewski
http://wpkg.org
______________________________________________________________________
OpenSSL Project http://www.openssl.org
User Support Mailing List [email protected]
Automated List Manager [EMAIL PROTECTED]
