> < One guess would be that your client is using ssh2 by default, and NetBSD
> < is accepting ssh2 connections. Try again with ssh -1.
>
> Using ssh1 connections does significantly speed up the connecting, but
> doesn't explain the original discrepency.
>
>         OpenBSD NetBSD
>          Q610    IIci
>
> ssh 1     6s      14s
>
> ssh 2    23s     435s
>
> 435 seconds does not fit.

I would agree that a thirty-fold difference in speed is definitely an
indication of something being wrong. Up to a fourfold speed difference
between a 68030 and 68040 at the same clock speed would be understandable.

My guess is that the machine that was used to make the binaries made
"generic" ssh binaries. For ssh, there's some assembly for the
cpu-intensive parts, and the "M680x0=YES" option in /etc/mk.conf allows
assembly for a particular CPU to be used.

For example, on the m68040, 64 bit multiply instructions are used; on the
m68060, these instructions are emulated, so using m68040 code would be
much slower than compiled c code.

On ports that have no assembly, generic c code is used; perhaps, for some
reason, the generic c code was used in the ssh in the distributed
binaries.

Can you compile with M68030=YES and, perhaps, COPTS+=-O3 in /etc/mk.conf?
If not, I can compile the binaries on one of my faster systems if you're
willing to test them...

John Klos
Sixgirls Computing Labs