Fedora target audience, what about the research area? (was Re: Leaving?)

Laurent Rineau laurent.rineau__fedora_extras at normalesup.org
Fri Jul 28 15:29:31 UTC 2006


On Friday 28 July 2006 11:37, Thomas M Steenholdt wrote:
> I'm firmly of the belief that Fedora should NOT wait for anything like
> this. Problems like these are best regarded by the 3rd party repos like
> livna or atrpms. However, I can understand the unfortunate problem for
> users who depend on the binary drivers too. My thought is that if an
> update causing problems for a non-free, binary-only driver is  not
> acceptable (it's acceptable for me and the proprietary drivers I have to
> deal with on certain Fedora machines), then they should probably choose
> a different platform for those systems (I have lots of RHEL systems too
> because of support issues with proprietary stuff too). There are several
> free and non-free projects that will be more ABI stable than Fedora (And
> there are free RHEL-like distros to choose from as well). But a lot of
> us LIKE this characteristic about Fedora. Actually all the catering to
> binary and non-free stuff are some of the things that keep me away from
> other distros for certain types of machines. I suspect a few other
> Fedora users feel the same way about this.

(Maybe we should have discussed of that in the thread "Fedora's intended 
target audience?". So I have at least changed the subject.)

Since Fedora Core 1 has been published, several French public research 
institutes have decided to switch to FC. For example:
  - the ENS ("École Normale Supérieure") of Paris, has switched from a non 
uniform park of FreeBSD/Redhat/Solaris machines to an almost full exclusive 
park of FC machines,
  - the Inria ("Institut National de Recherche en Informatique et en 
Automatique") of Sophia Antipolis has switched from a home-maintained 
Redhat-6.1 (with a huge /usr/local on NFS) to Fedora Core.

In both cases, the reasons were:
  - easier security updates,
  - easier way to maintain a system up-to-date (but not bleeding-edge, in 
which case they would probably have switched to a Debian-testing),
  - free (as in "free beer"): these institutes could not decide to buy RHEL, 
which would be to much cost for the whole park of machines. The budgets in 
public institutes are decreasing each year.

At the ENS, it is 258 machines, and at the Inria it is 689 machines.

These ABI breakage we are talking about (soname clashes, kernel changes, x.org 
ABI) are a pain in the ass of the system administrators, for several reasons:
  1/ Every computer science projects, in these institutes, are developping 
software with Fedora as the developping platform. Some of them are even 
developping free software. Here is a list of free software packages developed 
at Inria, for example:
  http://pauillac.inria.fr/cdrom/prog/unix/eng.htm
Some of them are packaged by Fedora, but some others are not yet ready for 
inclusion in a linux distribution. Each time a library changes its soname in 
Fedora, and no compat-* package is published instead, then some tools 
developed locally need to be recompiled, if possible, or the old version of 
the library has to be installed in /usr/local/.
  2/ Some binary kernel of X11 drivers are needed, because the researchers of 
these institutes do no always check the support status before buying a new 
laptop, for example. Or because some research projects have a real need of 
3D acceleration (when dealing with surface meshes with millions of 
triangles). The kernel issue is handled by the administrators: they use their 
own home-made and patched kernel. But the X11 problem would really be a 
problem. The administrators recommend to researchers to "yum update" their 
Fedora regularly. If you decide to push x.org-7.1 is FC-5, the next yum 
update will lead to X11 servers not run at all!

In one hand, it seems normal that a FC major release (FC-6) breaks ABI 
compatibility. One the other hand, administrators and researchers expect 
Fedora Core not to be "broken" by daily updates.

I understand those who answer "Fedora will not be broken". Yes, that's true. 
But the ABI will be broken for third party software, even if they are under 
GPL! And the X11 servers of a big part of the Inria will crash with next 
daily update.

It would be nice that those ABI incompatibilities are avoided during the life 
time of a major release of FC+FE. If people wants bleeding edge software, 
they can use FC+FE-devel, or special third part repo that will backport some 
of the FC-devel software to FC-5.

If FC is "broken" too often, from *the user perspective*, some users will just 
switch to another distro, and I do not think it will be good for Fedora.

As regards the institutes I told about above, if some ABI breakage occur too 
often in the life time of a major release of FC+FE, I know that the 
administrators will switch to something else. I hope that we will avoid that.




More information about the fedora-devel-list mailing list