Solaris is significant faster in disk performance on a normal SCSI
disk through the filesystem. These are numbers for the same disk moved
between machines, using the bonnie benchmark (see my home page also).

empire (SS1, 24 MB), Solaris-2.5
          140   181 97.7  1391 61.0   395 28.3   134 98.6  1988 86.5  49.4 20.6
          140   181 97.7  1376 60.5   421 30.1   134 98.7  2024 85.5  51.1 21.1
grisu (SS1, 20MB), NetBSD-1.1
          140   200 97.2   798 23.1   458 26.1   137 97.4  1353 41.6  44.4 11.6
keks (SS2, 32 MB), SunOS-4.1.3_U1
          140   381 98.1  1705 32.4   447 17.7   266 97.6  2258 49.3  54.5 12.3



I like to run a little paing benchmark, probably far away from any
application. While they are nearly equally fast, NetBSD gets really
inresponsive compared to Solaris while this test is running. NetBSD's
VM system is really in the way here, an effect that shows best on i386
compared to FreeBSD's VM system performance.

empire (Solaris 2.5, 24 MB RAM, 27 chunks)
gesamt: 4.0u 47.0s 11:13 7% 0+0k 0+0io 0pf+0w

grisu (NetBSD-1.1, 20 MB RAM, 22 chunks)
gesamt: 1.4u 28.4s 10:55.23 4.5% 0+0k 553+10io 0pf+0w

Another datapoint: The machine is under load, 5x dhrystone and 20x
tail -f (the tail -f just to test for OSes that put idle processes in
false queues). No, how long does it take to get the answer from `rsh
box-under-load ls /`? Di the repeated, will it speed up next time?

grisu (SS1, NetBSD-1.1) 20 MB:
0.00user 0.03system 0:03.59elapsed
0.00user 0.03system 0:02.80elapsed
0.00user 0.03system 0:03.10elapsed
0.00user 0.03system 0:03.48elapsed

joker 486 DX-50, 8 MB netbsd-1.1:
0.00user 0.03system 0:07.19elapsed
0.00user 0.03system 0:01.35elapsed
0.00user 0.03system 0:00.44elapsed
0.00user 0.03system 0:02.50elapsed
0.00user 0.03system 0:04.76elapsed

empire (SS1 24 MB, solaris 2.5), simulierte dhry
0.00user 0.03system 0:09.56elapsed
0.00user 0.03system 0:09.70elapsed
0.00user 0.03system 0:11.48elapsed
0.00user 0.03system 0:07.64elapsed

Here Solaris really loose. I have heared two explanations for this
effect:
- when a certain number of processes that are ready to run (dhry is
  always ready to run) is reached, Solaris will slow down task
  switching. 
- Solaris exessive use of shared libraries will make process
  invocation (this benchmark doesn't do much else) slow.
Much more, ask 4 people, get 5 different answers. A real reason to
prefer a system with source.

All in all, I found NetBSD to be faster than Solaris for running
normal commandline Unix-work and forking network services that don't
stress things too much.

I found Solaris to be faster when it come to run nothing than one
specific application is one precess tthat expecets to be fed from the
filesystem. 

It seems both systems are better in those areas they're designed
for. Surprise. Both systems are unacceptable when you overstress
memory, Solaris just wants more memory, NetBSD's VM system can't make
real use of memory.

Those of you how found NetBSD (or Linux) to be two times as fast as
Solaris: I don't beleive you, could you please be more specific what
exactly you did? Ideally, post an URL to whatever program is that
fast. The only situation I could imagine where this can be true is a
heavily forking shell script. I don't consider this to be an important
benchmark to choose a system for commercial application load.

Don't understand me wrong: I like NetBSD. But I hate false or for that
matter unproved claims of being superiour. Sounds too much like most
of the fanatic Linux people selling Linux as beeing able to run
Windows applications, in wine. (This is not a comment about serious
Linux folks).

Martin
-- 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Martin Cracauer <cracauer@wavehh.hanse.de>  http://cracauer.cons.org