Re: trimmed down rescue/root/driver disk sets
Randolph Chung <firstname.lastname@example.org> writes:
> I am working on trimmed down rescue/root/driver disk sets. With Adam's leave
> I am investigating the effects of using custom kernels instead of the stock
> kernel packages we have in the archive. On i386 at least, this can result in
> quite significant space savings.
> On i386, preliminary tests show that I can create a root/rescue/driver set
> with support for common setups (ide, scsi, pci/isa net cards, pcmcia, but no
> raid) in 3 disks (1 root + 1 rescue + 1 driver) instead of the current 5.
> It'd be really nice if the root and rescue image can fit on the same floppy,
> but i don't know if that'll possible. The kernel image I have now is about
> 940k. :(
Good -- what about your investigations on unnecessary kernel modules
for *all* flavors? By unnecessary, I mean anything not necessary for
installation. This includes sound drivers, IP masquerading modules,
IPv6, and the like... I was hoping we could get down to 2 drivers
disks on i386 for the "kitchen soup" flavor.
> A question for you all: what is the minimum set of options that need to be
> built into the kernel? How much of it can be built into modules and still
> allow for easy installs? In my experiments, I am building in some common pci
> net cards (again, for i386) -- tulip, via-rhine, eepro, vortex, etc. I think
> this will work well for people who plan to use Marcel's net-install code.
> can we do anything with scsi support? can that all be modularized without
> sacrificing ease of install?
Let me redirect your investigations a bit. You should provide these
alternate kernels/drivers as i386 "flavors".
I'd like to remind people that it is *only* i386 which doesn't provide
stock kernels with networking support for 90% of platforms! I
*refuse* to complexify boot-floppies to accomodate the hardware
nightmare of i386.
I would suggest you *not* bother trying to go back to one disk for
rescue and root. The kernel size is too big.
It is too late in the process to have an initrd scheme, which is has
the best possibilities for user convenience. It's just too much to do
and we haven't the time.
That being said, remember the issue is *convenience*. Suppose we
could provide a kernel-image-2.2.13-pci-ide and
kernel-image-2.2.13-pci-scsi packages. Those should include *common*
IDE or SCSI drivers and common PCI (or whatever) network drivers.
Remove old legacy crap such as FPU (387) emulation, MCA support, and
Then we could provide scsi-pci and ide-pci i386 *flavors*, that is,
additional rescue/root/driver disk sets for i386. Ideally, 80% or
better of *new* computer buyers could proceed to use these rescue/root
disks to boot into a network-capable system, and then they could
install the rest of the system (via ethernet, ppp, whatever) from a
local server or even from http.us.debian.org.
Thus we would have a scheme where they can use the kitchen-sync option
with the 2 or 3 drivers disks for installation, or they could use the
scsi-pci or ide-pci rescue/root only and install the rest from the net
(no drivers normally needed).
I really suggest you look at Tom's Boot disk for inspiration, esp.,
the way he's got the kernel compiled and what options are added in.
You're going to have to *first* get the kernels built and uploaded for
pci-scsi and ide-scsi, and ideally those should be in Debian itself as
well, so that's the first place to start.
.....Adam Di Carlo....adam@onShore.com.....<URL:http://www.onShore.com/>