[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Stop archive bloat: 47MB gmt-coast-full_19991001-1.deb



On Mon, 18 Oct 1999, Jason Gunthorpe wrote:
> On 18 Oct 1999, Philippe Troin wrote:
> > Do we really need a 47MB file that will be useful to maybe 10 people
> > using Debian (at most) ? We already fill 5+ CDs...
> On the same subject, the Debian archive is now 10.8G in size, at the
> current rate of growth a majority of the mirrors will likely stop
> mirroring non-i386 by the end of next year and if things continue in two
> release time we will be like 20G big and we won't even be able to mirror
> it ourselves!

The problem that we face compared to a model like Windows or MacOS is that
we have the operating system plus *all* application software housed at the
same place. That means we necessarily have a huge archive, if we want the
same kind of functionality.

I don't know how the commercial *nixes handle this, but it seems they
distribute things differently from us too.

We have advantages over the Windows and Mac models, though. We have shared
libraries and dependency information that all works much more smoothly
than anything they try to do. Just take a look at how many *.dll's each
installer for Windows includes. Our system for installing and uninstalling
packages is much smoother too, if not as idiot-friendly (I think it is
user-friendly to a large degree, with room for improvement).

So here's my idea. I think we should leave only the stuff that makes up
the basics of the OS in main. This would include all of the current
Required priority, probably all of Important, probably all of Standard,
plus things like X (everything needed for a running X environment,
possibly including (part of) GNOME or somesuch), and sections like:

o base
o admin
o relevant parts of doc
o maybe part of editors
o some or all of libs/oldlibs
o perhaps part of mail
o parts of misc that have better places to go (cbb, kernel-package, screen)
o part of net
o some of sound
o most/all of text
o part of utils
o part of web
o parts of x11 (see above)

Then we distribute that as one CD set. This would generally include
everything you'd expect to find on a commercial *nix distribution CD,
maybe pruned a bit (my educational copy of Solaris 7 fits on 1 CD +
1 more of documentation).

Next we create the data section according to current policy, along with
creating a new section (or maybe a few new sections) for applications.
Data would include what's already been discussed. Applications would
include things like:

o WINE
o XEmacs, other emacsen
o maybe a good portion of GNOME (if not all)
o all of games
o most of devel and interpreters (all the -dev pkgs, all the stuff not
  necessary for compiling downloaded source. This might also go in a
  separate development section)
o all of graphics
o parts of libs that aren't required by stuff in main
o math
o most of mail
o a fair portion of net
o some of sound
o all of tex (teTeX and all the fineries)

Then we distribute basically each section as a separate CD set, with the
option to get all of them.

There's several hurdles to doing this:
o It would require major policy changes.
  o Perhaps the most difficult part.
o it would require significant, though not major, changes to
  dselect/apt/console-apt/whatever.
  o This would mainly include just UI changes, to print sections prettier.
    console-apt currently only shows base, admin, etc., not non-US/base, etc.
o it would require changing the sections of a lot of packages, and even
  moreso changing priorities of a lot of packages.
  o This would be a major hurdle to overcome.
o it would require an increase in hardware capabilities by our servers
  (maybe mirrors too -- but they could just mirror one section each if
  they wanted)
  o Maybe we start apps.debian.org, main.debian.org, data.debian.org to
    handle each section. Maybe even set them up as a round-robin system of
    mirrors.
o it would complicate our distribution methods and our installer.
  o Maybe have each CD have its own installer.
  o The multi_cd method (of dselect or apt) would have to be smooth (I
    don't see them as real smooth currently).

But there's advantages:
o it would enable us to keep packaging everyone's favorite software
  packages for debian to make software accessible to anyone
o it would make it easier for people to say "I want these components for
  this system" and buy CDs for just those components, without having to
  get stuff they didn't want just laying around.
o It makes more sense to me to do it this way -- better organized than
  currently.

This would definitely be an eventual goal, but I release the idea for
discussion. I'm aware of many of the downfalls of actually doing it, but
feel free to discuss it, reject it if you so desire.

--Kevin R. Bullock

kbullock@ringworld.org | TheMystic@EFNet
krbullock@csbsju.edu   | Caoimhin@EFNet

"One World, one Web, one Program" - Microsoft promotional ad 
"Ein Volk, ein Reich, ein Fuhrer" - Adolf Hitler 


Reply to: