1 |
Hi guys, |
2 |
|
3 |
I realize that this is a bit late to bring this up, but I wanted to |
4 |
suggest an optimized organization for our x86 CDs. |
5 |
|
6 |
Traditionally, we've had 5 disc 1's and 5 disc 2's. What I'm proposing |
7 |
is that we move to 1 disc 1 and 5 disc 2's. There would be one disc 1 |
8 |
for all our sub-architectures. This would help the store in that we'd |
9 |
have fewer unique CDs to produce. It would also remove about 2GB from |
10 |
the mirrors. |
11 |
|
12 |
Here's how disc1 would be organized (I'm using shorthand names for the |
13 |
files): |
14 |
|
15 |
(runtime data, about 75 MB) |
16 |
/stages |
17 |
x86-stage1.tar.bz2 |
18 |
x86-stage3.tar.bz2 |
19 |
i686-stage3.tar.bz2 |
20 |
pentium3-stage3.tar.bz2 |
21 |
pentium4-stage3.tar.bz2 |
22 |
athlon-xp-stage3.tar.bz2 |
23 |
/snapshots |
24 |
portage.tar.bz2 |
25 |
/distfiles |
26 |
225MB of sources (everything needed to go from |
27 |
stage1 to stage3 and to boot the system, plus kernel sources, |
28 |
kernel-dependent ebuild sources, and if there's room, X sources. |
29 |
|
30 |
Necessary changes to the docs to accomodate this: mention that if the |
31 |
user wants to install from stage2, the stage2 will need to be grabbed |
32 |
from our mirrors using wget. Most users don't use stage2, so this won't |
33 |
impact very many people. |
34 |
|
35 |
Users no longer need to do a cp -a /mnt/cdrom/packages/All/* step for |
36 |
CD1. This simplifies installation. The downside to this is that |
37 |
things like metalog and vixie-cron will need to be compiled from |
38 |
source. This isn't too horrible since they don't take particularly |
39 |
long to compile. And the sources for these packages will be included |
40 |
on CD1. |
41 |
|
42 |
The lack of .tbz2s on CD1 would also mean that for the entire |
43 |
installation process up to the first reboot, you don't need to use |
44 |
"emerge -k." But using emerge with a -k would still continue to work. |
45 |
This could allow for simplification of the docs (remove unnecessary '-k' |
46 |
references.) |
47 |
|
48 |
All in all, this would require very few changes to our existing install |
49 |
procedure. No significant changes would be necessary. |
50 |
|
51 |
CD2 would contain pre-built packages (the *entire* GRP set) for a |
52 |
particular sub-arch, such as athlon-xp. The user would be able to pull |
53 |
from the GRP set after he/she reboots the box into his new install and |
54 |
mounts CD2. |
55 |
|
56 |
I think this new layout is a good idea for several reasons. First of |
57 |
all, it helps the store to deal with our continually increasing order |
58 |
level, which will probably skyrocket with the 2004.0 release. I will |
59 |
need to outsource the production of at least some of our CDs, and doing |
60 |
1 run of 2500 CDs is a *lot* cheaper than doing 5 runs of 500 CDs. |
61 |
|
62 |
Second, it takes some load off our mirrors, and gives users only one CD1 |
63 |
to download for all of x86, and the CD1 that they get is a lot more |
64 |
versatile than what we currently offer. The 2GB freed up on the mirrors |
65 |
could be used for the upcoming x86 KDE/GNOME LiveCD, as well as one or |
66 |
two other special catalyst LiveCDs. This allows us to offer more cool |
67 |
stuff to our users without adding any additional bloat to our mirrors. |
68 |
|
69 |
Likewise, if this is adopted for PPC, it would allow us to have |
70 |
potentially 1 instead of 3 (or maybe even 4?) disc 1's, and reduce our |
71 |
mirror load by an additional 900MB to 1.4GB. It would also give ppc |
72 |
users one CD1 to download that provides them with pre-built stage3's for |
73 |
ppc, g3, g4 (32-bit g5?) which I think many would find convenient. |
74 |
|
75 |
Please let me know what you think of this new layout proposal. |
76 |
|
77 |
Regards, |
78 |
|
79 |
Daniel |
80 |
|
81 |
|
82 |
-- |
83 |
gentoo-releng@g.o mailing list |