[sclug] Gentoo...

[Dickon Hood]

On Sun, Nov 28, 2004 at 13:25:24 +0000, Alex Butcher wrote:
: On Sun, 28 Nov 2004, Simon Huggins wrote:
: >On Sun, Nov 28, 2004 at 11:51:09AM +0000, Darren Davison wrote:
: >>On Sunday 28 November 2004 11:34, Hamlesh Motah wrote:
: >>>Anyway, my question is simple, why has there been so much hype over
: >>>Gentoo recently?

Because they're gaining lots and lots of users, and evangelising a lot.

: >>The ability to compile all software to match your own hardware (rather
: >>than the lowest common denominator that pre-compilation necessarily
: >>dictates) is great

: >What do you think you gain from this specifically?  Do you mean in terms
: >of features you turn on/off (e.g. picking one of OSS or ALSA) or in
: >terms of compiler optimisations?

: Indeed; once you've got an optimized kernel, glibc, and crypto libraries,
: you're rapidly into diminishing returns. I'd be amazed if anyone even
: recoups the compile time from improved runtimes.

And there's certainly nothing stopping you doing that anyway when using
one of the 'binary' precompiled distributions: just download the source
.debs / .rpms (your poison of choice) and recompile with all the
optimisation flags you want.

I did this quite sucessfully with xine's rendering libraries, partly to
get some performance improvements out of it (I'm trying to run it on a
900MHz Athlon; it's a bit marginal when I've got webbrowsers and whatnot
on it too), and also to install some patches into it to support my
PVR-of-choice software.

The kernel is the one thing I don't agree with using precompiled
distribution versions of.  IME Redhat in particular are bad at applying
patches which aren't in the stable trees, notionally to fix bugs, but
often to add features.  The device driver selection is generally not
optimal, too.  libc is something I also avoid touching, but mostly because
it's just messy to upgrade, and horribly trivial to wreck your system
doing it.

: Additionally, i386 RH and Fedora are optimized for i686 (or Athlon, if you
: install the .athlon.rpm packages) but only use the i386 instruction set, so
: the binaries still run on strictly-speaking-686-but-not-quite-in-practice
: CPUs such as some of the VIA chips. Again, the performance gains from being
: able to use CMOV and such are relatively insignificant.

With the P4-class CPUs, instruction scheduling plays a *very* significant
part in performance, rather than just the added instructions you have
available.  Interestingly, there are some cases which Intel have
introduced in the P4 that are pathological in the i386, and vice versa...

'Gentoo users' computers don't run slowly because they're running
unoptimised software.  Gentoo users' computers run slowly because they're
constantly recompiling everything.' -- I forget who said that.

-- 
Dickon Hood

Due to constant nagging to change it, my .sig is temporarily unavailable.
Normal service will be resumed as soon as possible.  We apologise for the
inconvenience in the meantime.