[plug] Distro Day !!!

[William Kenworthy]

Random thoughts ...
Ok, here's where it gets muddy!

One of Gentoo's strengths is customisability.  For instance, there are
31 flavours of kernel available - none is termed "THE" default kernel,
though gentoo-sources is a (very heavily patched) suggestion in the
install docs, along with an XFS patched one for those wanting that
file-system. This kernel has low latency and pre-empt patches, as well
as a kernel timer override for snappier desktop response. Other features
that can be added include the ccache and distcc features I have
mentioned previously, and object-prelink as an option for the OS as well
(I have not used this - yet!).

These are not "user add-ons", but true features of the operating system
with support and documentation built in.  Thus, for a developers
workstation, ccache and distcc would normally be installed, possibly
with a kernel to match the development target, but for a desktop,
obj-prelink, and all the desktop options would be used.  You choose the
task, and provision accordingly.

The more I look at this, the more a i386/i586/maxOpt comparison seems
difficult  For instance, I think latest Mandrake uses obj-prelink, but
RedHat doesnt (what does debian offer?).  If this is regarded as fair,
then should Gentoo be tweaked for maximum performance using all the
features offered by its customisability?  Gentoo is not "pick a lowest
common approach", but tune or retune for the task at hand, which will
make this more a debian/Mandrake/Gentoo comparison.

If you want to do a 386/586/maxOpt, you may be better using Gentoo,
compiled for each architecture to create a level playing field.  Maybe
as a followup?

For tests:
OO is often timed as to how long it takes to start from a) a clean
start, and b) restart.

Gnumeric is one program that wastes much of my time - it is easy to
create spreadsheets that take 20 mins or more to load - but often are
too big for OO, or excel (excel is much faster than either, until it
crashes, and burns your data in the process!)!


Athlon's/P4 - I have seen somewhere that athlons handle unoptimised code
much better than P4's, which benefit far more from optimisation.  An
interesting fact if true ... depends on what the test machines offer. 
Note that some athlon and p4 gcc options can produce invalid code
(sse2?), though this has supposedly been fixed in gcc3.3, which is not
in gentoo stable yet, as there a lot of problems with many ebuilds and
this version.  Reccomended for the P4 are the P3 opts at the moment -
note that this is a gcc problem, not Gentoo.  There is also an Intel
compiler option to replace gcc (already part of gentoo, a one line
command to install it, and a key to add to the environment to enable
it!) - some are reporting 15%-20% gains using this in some applications.

rattus linux # emerge icc -s
Searching...   
[ Results for search key : icc ]
[ Applications found : 6 ]
 
....

*  dev-lang/icc
      Latest version available: 7.1.006
      Latest version installed: [ Not Installed ]
      Size of downloaded files: 61,920 kB
      Homepage:   
http://www.intel.com/software/products/compilers/clin/
      Description: Intel C++ Compiler - Intel's Pentium optimized
compiler for Linux



BillK

On Sun, 2003-07-06 at 15:11, indy at THE-TECH.MIT.EDU wrote:
> Just to point out, I don't care which we use for the test.
> 
> It's a general performance test where we hope to discover
> some things about the benefits of Gentoo "optimized compiled"
> software, vs. Mandrake's pentium tuned packages and Debian's
> i386 packages.
> 
> To that end, in the DE environment is only really a factor
> in the "user feel" portion. We will be doing some hard numbers tests
> with Gimp and I'm still hoping someone can suggest some OOo tests,
> but since all three distros will be tested with the same desktop, I'm
> not sure it matters.
> BUT if someone knows a reason it does, I'd like to hear about it...
> 
> We won't be testing "ease of use", just feel of snappiness.
>