On Wed, Dec 02, 2020 at 04:25:38PM -0500, Josh Boyer wrote:
On Wed, Dec 2, 2020 at 4:18 PM Paul Bolle <pebolle(a)tiscali.nl>
wrote:
>
> Paul Bolle schreef op wo 02-12-2020 om 21:30 [+0100]:
> > Currently there seem to be over 6000 texlive packages. (Quick and dirty
> > measurements, sorry.) So splitting the kernel into an absurd number of
> > packages for (obscure) modules isn't a no-no on principle.
>
> If I sort the installed packages on my laptop by size the kernel-core packages
> don't even make it into the top 10.
>
> (It's a fun thing to do queries like that. glibc-all-langpacks is over 200M. I
> wonder what it does. Now I know that iwl7260-firmware is 125M! Linux firmware
> is almost 300M. Does Fedora support systems that are usable without it?)
>
> I'm guessing kernel-core and the various kernel-modules* packages total about
> 120M. So why not dump everything in a single package?
The larger your installation is, the larger the problem space you
have. For things like VMs where you want to upload/copy them around,
you incur transfer times, potentially networking costs, etc. At
scale, small size increases add up to large costs.
Also, that 'problem space' also includes the fact that it increases
the surface of attack. Some modules are auto-loadable by users (say
DCCP). If they have a vulnerability, it would be better if they
weren't even installed at all. One can always blacklist it and so, but
the management of that gets quite complicated. If you have a system
monitoring for known vulnerabilities, you would get warnings for
things you don't really use (unless you filter them out, but then
again, why have it installed?).
Fedora hasn't exactly taken off in the cloud or in a hyperscale
fashion, but that doesn't mean we shouldn't try to address these
things. The Minimization objective is a larger effort around what
initially started the kernel-core/kernel-modules split to begin with.
josh