Subject: Re: openssh on 1.4.2/m68k
To: Lars Hecking <email@example.com>
From: Ignatios Souvatzis <firstname.lastname@example.org>
Date: 03/30/2000 10:46:41
On Wed, Mar 29, 2000 at 11:52:00PM +0100, Lars Hecking wrote:
> > > Say, I have a ppp connection and start ssh from an xterm. X freezes
> > > for about 25 seconds, and when it comes back, the load is > 2.0.
> > > This is sort of extreme.
> > > Any explanations? Plain telnet doesn't do this. openssl is built
> > > with -O3 -fomit-frame-pointer by default, could this cause problems
> > > on m68k?
> That's not the problem. I recompiled with -O2 -fomit-frame-pointer, no change.
> > I bet the real difference turns out to be in ssh's Big Number library
> > (in m68k assembly) vs. openssl's m68k assembly. It might be worth
> > looking into. In the meantime, try blowfish instead of 3des.
> I'm using blowfish right now. Down to 24 seconds :-/
> Maybe that 68040 cache related fix in 1.4.2 wasn't such a great idea
> after all ...
Um, the only difference this should make is _incorrect_ vs. _fast_ results.
Of course, there could be a bug in there.
I guess this is unrelated:
- a fast pentium user complained, too, although "slow" is faster there
- the only other complaint about 68040 speed was about _exec()_ speed...
I hope openss*() does not exec dc to do each of the calculations ;-)
I guess I can do some tests with different OS versions later this week. If
I survive this cold, that is.
* Progress (n.): The process through which Usenet has evolved from
smart people in front of dumb terminals to dumb people in front of
smart terminals. -- email@example.com (obscurity)