I think we're talking cross purposes!
I was trying to say that if lilo could do RAID( 1 ) then you could have a
5(say) way mirror for the 10meg /boot partition... then lilo would be on
all the disks... then you could use all 5 disks in a raid5 for /
On Fri, 5 Feb 1999, m. allan noah wrote:
> what you mention is not possible for the reasons i outlined on non-raid1.
>
> under raid 5- the /boot directory inside the root partition does not exist
> on everydisk. it is striped across multiple disks with a parity stripe on
> one. lilo would have to pull blocks from different disks, and if one was
> missing, it would have to use the parity stripe to re-generate the missing
> data. to difficult, and prone to one major failure: suppose the mbr on all
> disks points to the map file using CHS addresses. this means that all
> disks that lilo needs must be accessable by the bios. (inside 1024 disk
> limit, on first two disks) the first chs address for the start of the map
> is on disk one, disk one dies. lilo never loads anything, cause it never
> starts.
>
> you must get the kernel into ram another way. i use floppies or 20meg
> /boot partitions on each disk, with a multi- lilo setup.
>
> al
>
> "so don't tell us it can't be done, putting down what you don't know.
> money isn't our god, integrity will free our souls" - Max Cavalera
>
> On Fri, 5 Feb 1999, A James Lewis wrote:
>
> >
> > Lilo being raid1 capable would be cool, it would be simple enough to have
> > a 10 meg mirror to have /boot on.... then / could be mounted from a raid5
> > or somthing... might even be cool to have the mirror across each disk in
> > the raid5 then you could guarantee it would find one and boot no matter
> > which disk had died! Mirroring the physical device (Rather than
> > partition) might also allow the lilo information to be mirrored
> > better....) but then if lilo were raid1 aware then this should not be a
> > problem... it should write a book block to all appropriate devices!
> >
> > What's the likleyhood of this happening? I guess one could stop the
> > mirror and run lilo against one of the devices... but that's less than
> > ideal to say the least!
> >
> > Also, I have not had a chance to test over multiple disks... but my
> > tests have demonstrated some good stability overall... I would love to
> > know when this will be concidered stable enough for merging with 2.3 or
> > even 2.2??? One thing I have noticed is that there are a few ways to
> > crash the raid code... try to init a mirror with only one disk... the
> > kernel daemon crashes and estimates an infinite amount of time to
> > reconstruct... you cannot then raidstop or sync or reboot!... This can
> > also be achieved (somtimes) by modifying a partition table on a disk that
> > has raid active on it.... even if you don't touch the raid paritions....
> >
> > Unfortunately I don't have the skills to help debug this, but it's the
> > only problem I can identify....
> >
> > On Thu, 4 Feb 1999, m. allan noah wrote:
> >
> > > ok- the main point with booting is thus- with the new raid code, you can
> > > start raid from the kernel without any utils on the disk. meaning, all you
> > > have to do is get the kernel into ram, and jump to its starting address.
> > >
> > > this is what lilo is for, along with loadlin, etc.
> > > they are needed cause the kernel, and esp. multiple kernels cant fit into
> > > the MBR on the disk. so the mbr contains basically the CHS or linear
> > > addresses of where the map file is, and thus the kernels. (simplified to
> > > the point of incorrectness)
> > >
> > > under RH, these all live in /boot, which is usually in the / partition.
> > > you have a couple choices.
> > >
> > > 1. mkbootdisk. works well, i use it on some. makes setup simple, no
> > > non-raid partitions, no multi-lilo.conf setups. but, it is the floppy, and
> > > they suck for HA. (of course if ha is your deal, you have multiple
> > > machines or multiple hardware raid adapters, etc)
> > >
> > > 2. small partitions at start of disk (within first 1024 Cyl) marked
> > > bootable, mounted as /boot /boot1, etc. this requires some fooling with
> > > lilo, etc, i used this on my latest box, works well (have floppy as
> > > backup)
> > >
> > > 3. modify lilo to understand raid. this would be non-trivial for all
> > > personalities except for raid1, as the actual kernel in those situations
> > > will be split across disks, requiring lilo to know how to read the raid
> > > superblocks, which is sometimes difficult, cause the superblock is at the
> > > end of the disk, outside of the bios 1024 cyl range lilo has to use.
> > > i had this working accidentally once. i setup lilo, somehow created an
> > > array (raid1) using the root disk, without disturbing the data. the
> > > kernels were
> > > in the same CHS locations, and lilo still ran, even though it was a raid
> > > partition. was impossible i thought, but neat! till i ran lilo again and
> > > it bitched, could not upgrade!
> > >
> > > the lilo option seems doable for raid1. hardware solutions are also
> > > feasable, but non-trivial
> > >
> > > al
> > >
> > > "so don't tell us it can't be done, putting down what you don't know.
> > > money isn't our god, integrity will free our souls" - Max Cavalera
> > >
> > >
> > >
> >
> > James ([EMAIL PROTECTED])
> > My operating system unders~1 long filena~1, and yours?
> >
> >
>
James ([EMAIL PROTECTED])
Vortex Internet
My operating system unders~1 long filena~1, and yours?
