Benchmarking Optimizations

Todd Pytel tppytel at
Tue Nov 12 17:34:52 PST 2002


Hi all,

There's been a lot of talk here lately about relative merits of
optimizations in different programs. And because of Gentoo, I hear
people in other forums discussing optimization as well. So I was
thinking I might try to systematically run some benchmarks, if possible.
I have a pretty good setup for doing this - I just finished RPM'ing LFS
4.0 (same idea as Jaco's LvR project, but slightly different style), so
it's easy for me to recompile any part of the system or the whole thing
with whatever optimizations are of interest, and cleanly swap different
versions in and out. I haven't done X yet, but I could probably have
that, Mozilla, and OpenOffice done in another week if I concentrate on
them instead of cleaning up the old packages. I've got a fast Athlon XP
machine and a slower Celeron laptop, and could dig up some older
hardware at work to play with if there's any interest. So, some

1) Well, to start with, what to run? I'm guessing that compile times
will depend mostly on GCC itself, which doesn't optimize well, so I
doubt kernel compiles or similar would show me much. What about
gzip/bzip2? Would they show up any differences, or would they be bound
more by disk speed than anything else? OpenOffice and Mozilla load times
seem like good candidates. I could check out page rendering in Mozilla
as well. And glxgears was mentioned here recently. Other ideas?

2) Besides the software itself, what other kinds of controls would be
necessary? These are quiet systems, so I'm not worried about background
processes. How about memory cache? Is there a way to flush the cache
between runs, or set up tests so that its effect is minimized? File
location on the disk should be unimportant, since I can reuse the same
partition easily enough. Other things I'm missing?

3) For comparison to compiler optimization, what would be some useful
kernel tweaks/patches to check out? I've really never followed the
various kernel mods out there, but I've heard many people say that some
of them have a significant effect.

Ideas and comments greatly appreciated...


Unsubscribe: send email to listar at
and put 'unsubscribe blfs-support' in the subject header of the message

More information about the blfs-support mailing list