Re: Software Raid 5

DRBD-SSI is good for HA storage consolidation and when it works
reduces SPOF versus dedicated NAS.  Try to look at mountpoints in
OpenSSI like a virtual service in that the cluster as a whole can
survive by transparent failover to any node (any initnode for rootfs)
that are connected to the same storage.

Roger

On 6/19/06, Brown, Larry <[email protected]> wrote:
> I don't know if DRBD was a way to simplify the task or if there are
> other reasons it was chosen.  I didn't make the choice to use DRBD so I
> can't answer as to why it makes any more sense than any other mechanism.
> AOE stands for ATA over Ethernet.  There is a kernel driver that sets up
> a drive as /dev/etherd/eX.Y where X is a shelf the actual drive is
> designated to be on and Y is the drive within that shelf you are
> addressing.  The communication for reads/writes is done over Ethernet
> and so you can have raid over the network where two drives are on
> different racks / power supplies etc.
>
> I'm assuming that if you can have root failover with directly connected,
> shared, drives that the AOE method should work as well, but was hoping
> someone would have experience with this.
>
> I just don't grasp how the cluster handles the loss of the system that
> made the initial mount when it goes away.  For instance, if there was a
> complete array stored in an initrd and that initrd was booted and 3
> primary units joined the array, could you pivot root over to a software
> raid consisting of 3 drives one on each box.  Then if it is possible,
> could you then turn off the system from which the cluster was booted and
> the cluster continue to run?
>
>
> Larry Brown
>
> -----Original Message-----
> From: [email protected]
> [mailto:[email protected]] On Behalf Of
> David Bo Jensen
> Sent: Monday, June 19, 2006 2:45 PM
> To: [email protected]
> Subject: Re: [SSI-users] Software Raid 5
>
> What is AOE devices?
> If  Raid 5 is possible as you describe it, then DRBD will not make much
> sense, would it?
>
>
> ----- Original Message -----
> From: "Brown, Larry" <[email protected]>
> To: "Ed King" <[email protected]>
> Cc: <[email protected]>
> Sent: Monday, June 19, 2006 7:40 PM
> Subject: [Norton AntiSpam] Re: [SSI-users] Software Raid 5
>
>
> > My apologies, I should have been more specific.  I have also come
> across
> > a number of posts touching on software raid.  I was referring to
> > inter-node software raid IE:
> >
> > md0
> > /dev/1/hda2
> > /dev/2/hda2
> > /dev/3/hda2
> > /dev/4/hda2
> >
> > Where loss of node 1 would not constitute loss of the cluster.  If
> this
> > is not possible with the md code that is in the kernel due to the
> > underlying architectural changes, is it possible to do this with AOE
> > devices?
> >
> >
> > Larry Brown
> >
> > -----Original Message-----
> > From: Ed King [mailto:[email protected]]
> > Sent: Monday, June 19, 2006 1:27 PM
> > To: Brown, Larry
> > Cc: [email protected]
> > Subject: Re: [SSI-users] Software Raid 5
> >
> > If you want the root to be fault tolerant, a standard debian
> > installation allows this (Sarge 3.1).
> >
> > We partition each drive with a small boot partition (~50MB) , some
> swap
> > space (1GB) and the rest being allocated for RAID.
> >
> > The installer is told that the first drive, first partition is mounted
> > as boot, the other primary partitions on the other drives are not
> > designated to be mounted (the debian installer can't have more than
> one
> > drive mounted to the boot point).
> >
> > The installer allows the creation of the raid, and we flag that as
> > mounting as the root filesystem -- it gives us a very large software
> > raid, and on the faster processors can actually give better
> performance
> > than hardware raids.  The boot partition never changes (unless we
> > upgrade the kernel) and is small enough so we use dd to copy it over
> to
> > the other unused partitions so all the drives are now bootable.
> >
> > Brown, Larry wrote:
> >> OK, I'm going to ask a stupid question.  I'm sure I must have run
> >> across this in the past but can't find any mention of it now...
> >>
> >> Is it possible to allow a small group of nodes to boot to an init
> node
> >
> >> where the init node is still running off initrd, mount their disks
> >> with software raid, and pivot root over to them allowing the root
> >> filesystem to be fault tolerant with software raid 5?
> >>
> >> Larry Brown
> >>
> >>
> >>
> >>
> >> _______________________________________________
> >> Ssic-linux-users mailing list
> >> [email protected]
> >> https://lists.sourceforge.net/lists/listinfo/ssic-linux-users
> >>
> >>
> >
> >
> >
> >
> > _______________________________________________
> > Ssic-linux-users mailing list
> > [email protected]
> > https://lists.sourceforge.net/lists/listinfo/ssic-linux-users
> >
> >
>
>
>
>
> _______________________________________________
> Ssic-linux-users mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/ssic-linux-users
>
>
>
> _______________________________________________
> Ssic-linux-users mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/ssic-linux-users
>