current livecd is based on minifo
Advantages
- each module mounts as an overlay of the previous
- no requirement for ramdisks and symlinks etc.
Issues
- mini-fo is _BUGGY_.
- complex
- maintenance
- divergence (hardware detection)
- different kernel
- different X
- limited architecture support
- rsyncablity
New design (for hoary):
writable ( mini-fo )
-> device mapper snapshots
- copy on write at block level
compression ( cloop )
-> want to use cloop
- currently doesn't play well with device mapper (wrong blocks, corruption, badness)
hardware detection ( knoppix )
-> switch to hotplug
build system ( morphix )
-> debian-installer & custom bits
- need to better handle netcfg w/o dhcp issues.
- other architecture support comes for (hopefully) free.
- addressed by debian-installer
kernel ( 2.6.7+patches vs 2.6.8.1 )
-> unifying: add cloop to stock kernel.
X ( mkxf86conf )
- Xorg config ubuntu packages (using debconf)
New Fun Stuff
- persistant data (use a persistant device for COW)
- arbitrary storage ("it's a block device")
- debian-installer customization
- combined liveCD and install media
Misc notes
SquashFS is a read-only filesystem, which violates the design criteria.
From Juanje Tue Apr 5 14:06:42 +0100 2005 From: Juanje Date: Tue, 05 Apr 2005 14:06:42 +0100 Subject: Squashfs, ro, and so on Message-ID: <20050405140642+0100@https://www.ubuntulinux.org>
Current System:
ext3 image (rw) + cloop (ro) = cloop image (ro)
cloop image (ro) + blocks dev (rw) + device mapper snapshots = semi-rw filesystem
Proposal:
Squashfs (ro) + dir (rw) + unionfs = semi-rw filesystem
Reasons:
Cloop = Buggy, slow, inefficient
Squashfs = Stable, fast, very efficient, it suports more architectures
Unionfs extras = you can mount directories, it's no necessary to be a bloks dev. That open a lot of posibilities more than with just blocks devices'