Subject: Re: Port performance comparison?
To: None <email@example.com>
From: Andrew Reilly <firstname.lastname@example.org>
Date: 08/29/1997 09:23:44
On 28 Aug, Leo Bicknell wrote:
> Platform Chip Rating NetBSD Rating Native OS Rating
> pmax x mips x-y mips x-z mips
> m68k a mips a-b mips a-c mips
> That way you can compare what the processor should be
> able to do vrs what NetBSD is able to get out of it, as well
> as how well NetBSD stacks up to the native operating system
> for the box.
I think that you'll find that "Mips" ratings (or SPEC or Byte) will be
close to identical across operating systems, because they're all
processor/memory benchmarks, and operating systems don't (generally)
have much to say about them. Maybe the math library and the compiler
What I /would/ like to see, though, is a list maintained by some keen
individual keeping track of (for example) slowest and fastest times for
something like worldstone (cd /usr/src; make clean; make world) measured
for each port of NetBSD, annotated with a description of the hardware
used. The "fastest" point for each port would obviously involve some
sort of dragster hardware (RAID arrays, buckets of memory, grids of GHz
CPUs), but maintaining results from minimal configurations that can run
the benchmark would be illustrative too. (How many weeks to compile on
a 16MHz i386, 4M RAM and a single 3600RPM IDE drive?)
Of course there's probably quite a bit of variability across everyone's
"world" sets, so to be fair, and make it a benchmark, some fixed list
of common code would have to be defined.
I'd really like to see the results though.
"The steady state of disks is full."
-- Ken Thompson