[Mageia-dev] i686 must be Pentium II ?

[Giuseppe Ghibò]

2010/9/26 Tux99 <tux99-mga at uridium.org>

On Sun, 26 Sep 2010, Giuseppe Ghibò wrote:
>
> > Centos4 IS NOT a modern distro. It is a LTS started in 2005 and so it
> > maintains 2005's original skeleton of kernel, gcc, glibc and X. That's
> FIVE
> > years old.
>
> I'm quite sure Centos5/RHEL5 would install and run fine on it too,
> Debian would almost certainly too, the point is when you don't
> install/use a GUI, Linux still can run fine on very old low end
> hardware.
>

kernel 2.6.18 is a lot different than 2.6.31/33. As I said I invite to RUN
the installation/installer from scratch (there is the dual-arch installer on
a CD) of 2010.1 or cooker in such hardware. And if successful, report the
memory usage without any service started apart the login ttys. We are not
using the installation tools of CentOS or Debian. NetBSD would have probably
even the tiny one since AIM of being still compatible to m68k hardware.


>
>
> > As example MDV 2007.1, which is 3 years old, was still very
> > usable and responsive on my P4/ATI (maybe not as much as stable with 3D
> > acceleration), but 2010.0 ISN'T.
>
> Again, you are missing the point, you are talking about desktop/GUI use!
> Computers get used for a lot of other purposes, not just desktop/GUI use!
>
>


>
> > That's why I in some way asked a survey of oldest hardware based
> > on own experiences.
>
> Well, my oldest hardware that I still have working is a dual cpu
> Pentium 233MMX (the original i586) with 384MB RAM (currently has 2008.1
>

in i586 we are not even using MMX. As I said 2008.1 is not 2010.1 but 4
generation distro behind. Since it's not LTS, you might try to upgrade to
2010.1.


> on it) and a VIA C3 (samuel2 core, i586 since it lacks CMOV) box with
> 512MB RAM which has mdv 2010.1 installed on it and works fine for it's
> purpose too (headless home server running 24/7 and only uses 10Watts).
>

what is the output of "cpuinfo" there?


>
> > Very interesting, but will such "industrial use" will be target for
> Mageia
> > (BTW, certainly socket 775 CPU will support SSE and SSE2...)? If they
> have
> > an ISA slot, I guess is for maintaning the compatibility with some old
> fancy
> > (and maybe custom) card, certainly not for an ISA ethernet card that can
> be
> > easily replaced with a cheap PCI one or the one on board.
>
> That was just one example, there are many other situations were you
> still find ISA hardware, especially in developing countries.
>
>
> > Not exactly. I'm not talking in just using -march=<something> but in also
> > pushing -mfpmath=sse -msse (and maybe -msse2) , which should be much more
> > than JUST 1-2% (1-2% is usually the benchmark tolerance)...
>
> AFAIK SSE will only help with media apps (mplayer, etc) and they do
> autodedect already anyway so in practice nothing is gained.
>

-mfpmath=sse would replace the x87 with sse. Of course for any CPU not
having the SSE would result in a segfault or illegal instruction report
rather than a drop of performance as in case of changing the optimization
but maintaining the backward compatibility.


>
>
> > being realistic I think it's a lot of work and there aren't the
> resources,
> > so a choice of the default flags should be done.
>
> Agreed that would be too much work for very little benefit, the default
> flags of Mandriva are just fine since they still work on i586.
>

If you are able to break the Page's law you are welcome. :-)
[http://www.appscout.com/2009/05/moores_law_meet_larry_pages_la.php]


>
> We could have some selected packages also as i686 (like MIB does and
> like the kernel already is), like for example all the media players, but
> making the whole distro i686 would break too many uses of it.
>

as I said you are forgetting two of the most important like glibc and
kernel. As I said it's not that difficult to provide such packages (even
optimized for VIA C3 and C7) but require a little bit more than rebuilding
with specifying --target=c3,c7,xxx in rpm building.

I also think that sometimes application due to poor cache (including ATOM)
would run faster when compiled with -Os instead of -O2...; we could
introduce it for a common .i386.rpm package.

Bye
Giuseppe.

In a challenge of better supporting legacy hardware, why not adding Mageia
super-Legacy no-desktop section? where we are doing exactly the opposite of
supporting newer hardware. E.g.:

- remove i18n support, only LANG=C
- optimize for tiny cache
- reduce the # of fonts installed
- no themes
- reduce # of things in /etc/profile.d
- no extra audio daemon support (pulse, etc) or even no audio support
-------------- next part --------------
An HTML attachment was scrubbed...
URL: </pipermail/mageia-dev/attachments/20100926/47ebffc2/attachment.html>