Hi,

On Sun, Jun 06, 2004 at 02:21:09PM -0400, Vicky Staubly wrote:

[...]

> On my network, I compared 3 systems, connecting via ssh v2. The
> 90MHz Pentium beat a 167MHz UltraSparc by something like 3 seconds
> to 6 seconds.

Maybe. That would reinforce my initial statement. But: The time
between pressing [Enter] on the client and getting the login-prompt
from the server (90MHz Pentium1) is nearly zero. Both systems are
running Debian GNU/Linux.

> The 40MHz Sparc (SS2) came lagging along at about 30 seconds.

Yes, I can affirm this time.

> They were all pretty instantaneous for ssh v1, probably under a
> second or two.

You do not want to run SSHv1. Use telnet.

> (Oh, and the same 90MHz Pentium was dead even with a 90MHz Pentium
> running Linux 2.4.21)

Umh, I used to connect hourly to a 90MHz Pentium running GNU/Linux
from a 333MHz PentiumII running GNU/Linux whithout any problems.

Both systems had a 2.4.x Kernel and SSHv2 only (v1 disabled).

> I understand the OpenSSH code has some x86 assembly in it, 

Does it really? Would be sad somehow....

And how comes SSH is running so much faster on RS/6000 / AIX?

I will have talkback with a friend who's in the OpenBSD developers
team (They also implement OpenSSH). I hope he knows.

> but I'm not sure if that's the difference, or if the x86 has a
> better instruction set for ssh (I seem to recall during the
> distributed RSA cracking contests that there was a difference in
> performance based on how the architecture handled the shift/rotate
> bits opcode).

According to a statement of a friend of mine (not the one mentioned
above), only the key-generation process should be faster on intel
processors.

cheers,
\end{kryz}

-- 
Q: How is "SunOS" spelled?
A: As one speaks it. With capital "S-O-S".