Subject: Re: difference in key calculations for ssh
To: Grey Wolf <>
From: Andrew Gillham <>
List: current-users
Date: 05/25/2000 15:42:26
Grey Wolf writes:
> Why is it that a SPARCstation takes so much longer to calculate keys
> in ssh than does an i386?

My guess would be for the same reasons rc5des is significantly faster
on Intel.  My understanding is that the Intel is much faster at certain
instructions (shifts?) that are heavily used by the encryption code.

> My ss5/170 takes disproportionately longer to calculate keys and
> challenges and stuff for ssh than the PII/300 next to it -- the
> SS5/170 takes about 30 seconds to respond with a password prompt, while
> the PII/300 responds fairly instantaneously.

I am not seeing anything on the order of 30 times slower.
time ssh <remote box> /usr/bin/true
p133 (1.4.2) -> sparc2 (1.3.3, 50mhz no L2cache): 
0.899u 0.059s 0:08.71 10.7%     0+0k 8+4io 0pf+0w

p133 (1.4.2) -> p133 (1.4.2)
0.915u 0.039s 0:02.35 40.0%     0+0k 0+1io 0pf+0w

So comparing the processor speeds, lack of L2 cache on the sparc, I am
not all that surprised by the numbers.
Your numbers though, sound horrible.  My sparc2 (50mhz) is kicking your
SS5/170's butt. :-)

Try comparing the following on the machines:
time ssh localhost /usr/bin/true

sparc: 4.843u 0.080s 0:12.63 38.9%     0+0k 0+5io 13pf+0w
i386 : 0.899u 0.058s 0:02.49 37.7%     0+0k 0+8io 0pf+0w

So, about 5 times slower.  Let's see, 2.66 times slower clock speed, then
some factor for no L2 cache, and a possibly faster set of instructions on
the Intel.  Hmm, not totally out of line I guess.

Andrew Gillham                            | NetBSD ist Affengeil.                     | Nachts ist es kaelter
I speak for myself, not for my employer.  | als draussen.