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

Re: booting from a raid0 device.



On Fri, Apr 16, 1999 at 05:07:43PM +0200, Nils Rennebarth wrote:
> On Fri, Apr 16, 1999 at 09:46:13AM +0100, Enrique Zanardi wrote:
> > Not really. We may load IDE or SCSI modules from the initrd, or from
> > floppy. The approach I'm thinking about is something like the following
> > (for i386, I haven't thought about other arches yet)
> > 
> > "floppy-booted" installations
> > 	one rescue disk with:
> > 	 - "bare bones" kernel: no IDE, no SCSI, just floppy
> > 	 - initrd with IDE
> A kernel with IDE compiled in is much smaller (~10k) than a kernel without
> IDE and a seperate module.

But owners of a SCSI-only system may like to have no IDE support on their
running kernel to save a few KBs of RAM. We try to use the same kernel
during and after installation.

> > 	(optional) one rescue disk with:
> > 	 - "bare bones" kernel
> > 	 - initrd with some drivers for some popular SCSI cards
> same here.
> If I understand you right, "initrd with some drivers" means an initial ram
> disk with several driver modules in the filesystem on that ramdisk.

Right. Altough an alternate implementation with modules on a modules.tgz
file on the floppy disk may be interesting too (as it let's the user
change the modules set without rebuilding initrd). But that means
including floppy support on the kernel, and that may be tricky on
SCSI-floppy systems...

> > "CD-ROM-booted" or "network-booted" installations:
> Does network booted mean: with an EPROM on the card? In this case I agree,
> they are the same type.

Right.

> I would like to see a third kind: A rescue floppy with
> - "bare bones kernel" in fact not even floppy needs to be included
>   just the rescue disk filesystem.
> - an initrd with as many network drivers as possible
> Now a server on the network holding the rest may be accessed, once network
> is set up.

Yes, that may be a nice addition.

> We could leave out System V IPC for the bootdisks, saving a few KB, but we
> need to install another kernel for the user's system then.

That may not be a good idea, see above.
 
> I'll try to build such a kernel (the 2.2.5 version) to see how big it will
> become.
--
Enrique Zanardi					   ezanardi@ull.es


Reply to: