Subject: ssh-keygen and CFLAGS testing
To: None <port-sparc64@netbsd.org>
From: Joel CARNAT <joel@carnat.net>
List: port-sparc64
Date: 07/07/2006 17:49:32
Hi,
I wanted to see if my CFLAGS (in mk.conf) were of any use.
So I compiled security/openssh and did a few ssh-keygen:
# clear; while [ true ]; do time
/usr/pkgsrc/security/openssh/work/openssh-4.3p1/ssh-keygen -t rsa -f
/tmp/null.key -N '' > /dev/null ; rm /tmp/null* && sync ; sleep 300; done
With CFLAGS set (-mcpu=ultrasparc -mtune=ultrasparc -m64 -mvis -pipe)
17.45s real 17.39s user 0.03s system
40.93s real 40.90s user 0.00s system
14.71s real 14.70s user 0.02s system
8.69s real 8.69s user 0.02s system
6.02s real 6.02s user 0.00s system
23.71s real 23.66s user 0.03s system
Without any CFLAGS set (no /etc/mk.conf at all), I got those results
50.80s real 50.75s user 0.01s system
43.97s real 43.94s user 0.01s system
31.94s real 31.90s user 0.01s system
7.05s real 7.04s user 0.02s system
26.44s real 26.41s user 0.03s system
20.99s real 20.96s user 0.02s system
what surprises me the most is the various values I get.
when I did some "dd" testing to check disk access, results were quite the
same inside a test set.
the machine IDLEs when not ssh-keygen(ing).
it's an U10 with 512Mo of RAM, running on it's default IDE drive and
running NetBSD 3.0-RELEASE (gcc 3.3.3).
here's a few things I wonder:
1. why results aren't closer (inside a test set)?
2. what's best (for a productive server), more time in "user" and less in
"system" (no CFLAGS) or the contrary (CFLAGS set) ?
3. is this test good enough to be run on -CURRENT and check gcc 4 value ?
TIA,
Jo