Cipher 'none'

[Martin Forssen]

On Sat, 14 Oct 2000, Damien Miller wrote:
> On Sat, 14 Oct 2000, Pekka Savola wrote:
> > I did similar tests on my P2/266 system to see how fast aes128-cbc and
> > rijndael128-cbc were.  These were conducted with a 10 MB data off
> > /dev/urandom:
> 
> You are probably better off using 10Mb hole, so disk IO and buffer 
> caches don't affect the results. This is what I use:

One has to be very careful when comparing results like this. There are
many factors wich affects the results a great deal. For example the OS is
very important in this case. I did some performance measurements between
different versions of ssh and one setup I tried initially was to run the 
following on one machine (linux with 2.2.12 kernel):

     ttcp -t  -->  ssh -L  -->  sshd  -> ttcp -r

(i.e. a test of portforward performance on one machine). This test got a
really awful result, a couple of magnitudes worse than I expected. I then
tried the same thing on a machine SunOS 5.6 and got the expected result
,i.e. two orders of magnitude faster than the linux one. The lesson I
learned here was that the client OS can really affect things (in this
case I suspect the scheduling code in linux sucks).

What I am trying to say here is that in order to be able to compare
performance results one must have identical environments (hardware and
software) in which the measurements are taken.

	/MaF

PS The one thing I did find out about ssh2 performance is that the window
sizes for the channel flow control are really important and openssh had a
bug which really hurt performance (this bug was fixed some time ago).