Re: New style raid support during install (was Re: kernel support, v0.36 v's v0.9)
From: Adam Di Carlo <adam@onshore.com>
> "Glenn McGrath" <Glenn.McGrath@jcu.edu.au> writes:
>
> > In summery, this is what would need to be done as i see it
> > - the rescue disk would have to be patched with raid v0.90 (2.5KB
larger
> > bzImage than if no raid support)
>
> Ok -- I'm forwarding this also to the kernel-image maintainer, Herbert
> Xu, for comment. Could this be modularized, btw?
>
Actually, i miscalculated, 2.5KB is just for one raid mode, to support all
raid
modes would make the kernel 16KB larger :(
If the boot image is just maxed out, what about spliting it to have one
basic boot
image aimed at compatability, and a second aimed at more advanced features?
I guess that might get a bit confusing though.
The kernel needs Linerar mode and/or RAID-0 mode built-into the kernel to
have boot support for them built into the kernel (as far as i can work out),
maybe it would be possible to have some built-in, some as modules.
The raid modules are 72KB (hsm.o linear.o raid0.o raid1.o riad5.o
translucent.o) so they are a lot bigger than as compiled into the kernel.
> > - binaries from the raidtools2.deb would nned to be in the base.tgz
(84KB)
>
> Yikes -- that's a big big problem. Is there any way we could shrink
> this? Possibly a raidtools2-base package?
hmm.. i didnt realise this would be a problem, when i said base.tgz above, i
should have said base2_2.tgz which is 16MB.
84 KB is this size of the binaries from the raidtools2 packages, they are
already
striped, so i dont see how they could be made smaller, also this isnt
counting the
docs and stuff from the raidtools2 package.
>
> > - The menu system would have to recognise /dev/md0 as a valid target
> > partition (if it doesnt already)
>
> Hmm... that's from utilities/dbootstrap/block_device.c I believe.
>
> > - Check /boot partition isnt on a /dev/md?
>
> Here you're trying to make sure that you have a boot partition which
> isn't RAID, since lilo doesn't understand RAID, right?
>
Yes
> > If its too late in the development cycle to implement new feature such
as
> > root raid support then so be it, but it will be good feature to have in
> > there, and its something i am quite prepared to spent a lot of time on
in
> > the coming months (although the next 3 weeks i will be distracted with
uni)
>
> Well, if we are freezing now, it is probably too late. However,
> there's no reason why we can't update potato boot-floppies even after
> release, i.e., a point release.
>
Yea, sounds good to me, seems my time is limited till the end of the month.
> I agree this functionality would be very desirable, btw. We might as
> well take whatever steps we can to do this -- are you willing to work
> on the boot-floppies source to make it so?
>
Yes, i am willing to work on boot-floppies, its not a good time at the
moment
as i mention above, but i have 3 months break starting december, so i look
forward to doing some stuff with debian and another project i have planned.
Im not actually a debian maintainer yet, but im sure if/when i have
something
to contribute it could find its way in somehow.
Thanks
Glenn McGrath
Reply to: