[CLUE-Tech] KISS session tomorrow

[Dave Anselmi]

David Willson wrote:

> I'd like to see some real comparison between the performance metrics of an
> un-kernel-optimized system and the same system after the latest kernel has
> been downloaded and tweaked.  Will I see better frame-rates in Quake2?  Will
> my programs compile faster?  How 'bout my web-pages and Samba-shares; will
> they be more responsive?
>
> And can you have those done by tonight?  :-)  kidding, kidding...

That sounds like a GMTKS* session.

In the compiling that I've done, optimizations seem to make little difference.
That's probably because much Unix code was written for <20MHz cpus.  If you have
a routine that takes .1 sec and you optimize the hell out of it and it now takes
.05 seconds, you notice nothing.  I'm talking about gcc's -O options.  Of course
a heavily loaded server is a different story, and if anyone has benchmarks that
say otherwise I'd be happy to hear about it.  I appologize that I don't have any
hard data to support this - someday, maybe.

There are other things that seem to be more important than compiler optimization
tricks.  gcc's -Os (optimize for size) is reported to do a good job making code
small - important when running something off a floppy.  Removing debugging
symbols from executables can reduce their size quite a bit which means less disk
I/O loading them.  KDE benefits from a technique called object prelinking that
improves its load time noticably.  Dynamic linking can also improve
performance.  Note that in most of these cases, it's time pulling the file off
the disk that we're saving (and Linux does a good job caching frequently used
stuff).

The bottom line is that hardware is fast and cheap, so optimizing isn't as
important as it used to be.  I have some opinions on kernel compiling too, but I
see that Dan mentions them in his talk so I'll save them for the meeting.

Dave

* Give Me The Kitchen Sink