Re: How does md(adm) work with fake-raid ?

["Jiang, Dave" <dave.jiang@intel.com>]

On Thu, 2013-07-18 at 23:35 +0000, Dmitrijs Ledkovs wrote:
> On 18 July 2013 23:03, Martin Wilck <mwilck@arcor.de> wrote:
> > On 07/18/2013 10:37 PM, Francis Moreau wrote:
> >> Hello,
> >>
> >> Sorry if the question is stupid but I'm a rookie in md things, but I'd
> >> like to understand the big picture here.
> >>
> >> I've been told to use mdadm whenever possible even if my raid is
> >> handled by the bios (fake raid) which use the ddf metadata format.
> >> (unfortunately it seems that I can't desactive this fake raid in
> >> favour of linux soft raid). It's RAID1 BTW.
> >>
> >> So my question is rather simple: in my understanding the bios is doing
> >> the mirroring, but when setting up the md device, linux (kernel or
> >> userspace, I don't really know) also handles the mirroring for RAID1
> >> personnality. Is Linux clever enough to see that the mirroring is done
> >> by the bios in my case ?
> >>
> >> Could anybody teach me the big picture ?
> >
> > Fake RAID uses a part of every disk to record information about the RAID
> > arrays. This is called meta data, and your BIOS uses it for setting up
> > the drives.
> >
> > Under Linux, first you need a low level SATA or SAS driver that detects
> > your physical drives, e.g. the ahci driver.
> >
> > md can then detect the DDF meta data on your disk just like the BIOS,
> > assemble the array(s), mirror the data, and do other RAID operations.
> >
> > Distributions can set this up automatically. Currently most distros
> > don't do this for DDF (they do it only for fake RAID using the Intel
> > Matrix (IMSM) format). For DDF, for historical reasons, most
> > distributions will setup a mapping using dmraid (device mapper based
> > mirroring). That will also basically work, but it isn't a
> > fully-functional RAID implementation such as MD. The magic to set up
> > either MD or dmraid automatically as disks are detected is hidden in the
> > distro's udev rules, and possibly in the distro's installer logic.
> >
> 
> There patches posted to debian bug tracker to enable using mdadm in
> the installer to assemble/setup IMSM/DDF raid arrays, and thus using
> mdadm.
> I haven't integrated those, but am planning to work on merging them soon.
> 
> At the moment dmraid is used by default for both IMSM/DDF on Debian/Ubuntu.
> 
> My experience with these fakeraid arrays is very limitted, and I'd
> want to enquire of proper migration strategies from dmraid to mdadm.
> Whilst looking at the udev rules, at the moment, i have disable
> IMSM/DDF from assembly in mdadm udev rules, because dmraid has a nice
> property of "activating anything it finds".
> I suppose having both mdadm & dmraid racing to activate those drives
> wouldn't be nice.
> 
> How would one migrate from dmraid to mdadm? I was pondering about
> drastic measures: patch out ISMS support out of dmraid, make dmraid
> package depend on mdadm and make mdadm activate ISMS drives by
> default.
> But that sounds harsh, as I wouldn't want to cripple dmraid package
> for those who still prefer to use it.
> 
> Are there distributions which switched to mdadm by default for ISMS? Suse?!

AFAIK for IMSM, Fedora, SLES11sp2, and RHEL 6.x all are using MDRAID and
have been for quite some times.