[Mageia-dev] Mirror layout

[Michael scherer]

On Sun, Dec 12, 2010 at 12:36:05AM -0500, andre999 wrote:
> Michael scherer a écrit :
> >
> >On Sat, Dec 11, 2010 at 08:16:33AM -0500, andre999 wrote:
> >
> >>>Not to mention that a ratio of 2 mirrors in the USA out of a total
> >>>of 25 seems rather odd, for something that admins do not care.
> >>
> >>2 of 25 PLF mirrors in the U.S.
> >
> >Technically, 1, since the other is down ( and should be removed from
> >the list ).
> >So a ratio of 4%.
> 
> Unless you are going to analyse what is down for the other distros,
> you should say 2 ± 1, that is 4 to 12%

Ok, when I say down, I should say "the domain no longer exist". It is just
not registered. Not "down and it will be up" later, but "down someone didn't
bother to pay the domain". Obviously, I should not assume that people 
check facts before telling me my numbers are wrong.

And since other distributions have various systems to detect this ( mandriva have one, 
fedora have one, opensuse too ), there is no need to touch to the number.
PLF do not have any checking tasks, so the mirror was not seen as wrong.

There is 1 working mirror and the svn changelog show that this number is quite
stable enough.

And I would have removed the incorrect one, if I didn't consider this as 
a abuse of my root privilege on zarb.org.

> Or 9%.  Depending on how you want to fudge the figures.

There is no estimate or fudging involved, we have exact number 
of mirrors, I gave the url for each distributions.


> But maybe it is because they (in policy at least) exclude non-free
> software ?

So does debian. 

> And just how rigorously do they apply a no patent-constrained
> software policy ?

A quick research could have answered to this question :
http://fedoraproject.org/wiki/Software_Patents

They used to remove mp3 support from source code :
http://www.csparks.com/redhatUnhoarked/index.xhtml

But that was 5 years ago. Nowadays, I do not think they still do it
as icecast for example is not modified ( despites supporting mp3 format
but maybe because there is no trace of codecs, it is ok ).

> Haven't I heard somewhere that Fedora (and RedHat) are based in the
> U.S. ?  So wouldn't it be natural to expect that it would have a
> higher proportion of sites there ?

Debian too is based in the US ( managed since 1996 by SPI, based in NYC ). 
So does Novell ( created in Utah, headquarters in Massachusetts ). 
 
> >And I didn't count other country such as Japan, where patents on software
> >are permitted ( http://en.swpat.org/wiki/Japan ), and where the count of PLF
> >mirrors vs Fedora mirrors is 0 to 8.
> 
> 0 ± 1 gives 0 to 12%.  Same ballpark.
> Also, recruiting Fedora mirrors could be driven by the commercial
> interests of RedHat.

"could" is a supposition, and I think you should give facts, not suppositions.

For the mirror, there is 2 private R&D labs ( KDDI, RIKEN ), 2 university 
( Yamagata, JAIST ), and the rest are network related ( iij.ad.jp, wide.ad.jp, 
dti.ad.jp, ftp.ne.jp see http://en.wikipedia.org/wiki/.jp for the meaning of 
the various second level domain ). 

So I doubt that commercial interest of the main sponsor have something to do, 
since the profile is quite similar to the usual one of most mirrors ( ie, 
people with lots of bandwidth, servers, and interest into helping free software ).


> >More ever, the fact that this is hosted by some private and rather anonymous
> >company is also a important point. Ie, no .edu or big telco ever contacted
> >PLF to host a mirror, while in France and another country, PLF have both.
> 
> Considering that PLF is based on Mandriva, and Mandriva is based in
> France, wouldn't it be natural to expect PLF to be better
> represented there ?

I think you missed the point. Let me explain :

There is no USA university, nor USA telecom company that contacted PLF.

On the other hand, in other part of the world, PLF is mainly hosted by telecom 
company ( like Zoomnet and Bentel, for example ) and by universities ( Porto, Taiwan, 
Bahcesehir among others ). 

> >>Also, there are only about 400 packages for i586 in PLF mirrors.
> >>Since most are duplicated, I wonder how many distinct packages there are ?
> >>Somehow doubt that an unlicenced copy of quotes from the Simpsons
> >>(one of the 2 plf packages that I didn't find also in Mandriva main)
> >>is going to be a big attraction.
> >
> >You should look a little bit more closely. For example, libdvdcss2 is plf only.
> >So does various emulator, lame ( and related like darkice ), gstreamer-bad,
> >etc. There is amule, and similar software. More than 2.
> 
> Of the twenty or so PLF packages that I found looking through
> available packages with Mandriva and PLF repositories enabled, only
> 2 did not also have the same version in Mandriva.  (All Mandriva
> main, in this sample.)  That is about 10% not in Mandriva.
> So for arguments sake let's say 20% are not in Mandriva.  That makes
> only about 80 packages only in PLF.
> Impressive, isn't it ?

You said on https://mageia.org/pipermail/mageia-dev/20101201/001576.html
that you have decades of programming experience. So I assume that writing
a script to get more precise numbers would not be too hard instead of 
saying "I counted somewhere 20 packages in a limited part of the 
distribution" :/

> BTW, gstreamer*plugins-bad is in Mandriva contrib.

But not all subpackages. Take a look at the spec file ( using 
mdvsys should ease the work ) and see that 5 subpackages 
are conditionnaly built.

I think you may have missed the point about PLF rpms being at Mandriva.
( or the contrary, depend on how you look ).

They share the same source code, but they do not link to the same 
software, or use the same configure options. 

So Mandriva considered that distributing mplayer without enabling mp3
write support ( with lame ) was safe enough. PLF do the distribution
of lame, and rebuild mplayer with it ( so mencoder can write mp3 ). 
Fedora do not distribute mplayer at all.

And that's basically the same scheme for various dual life packages 
( with variations about the feature that is enabled, there used to be a issue
on font hinting, for freetype and bytecode interpreter )

-- 
Michael Scherer