On Apr 8, 2009, at 20:10 , David Young wrote:
On Mon, Apr 06, 2009 at 11:26:19AM -0400, Andreas Wrede wrote:On 06-Apr-09, at 02:18 , George Michaelson wrote:is GPT mature enough, that software raid on GPT is viable? Because I am very very very scared of being exposed to 'you are locked to thisraid card, this physical order of disk' problems, and GPT sounds likein principle, it avoids thatI just rebuilt a backup server with a 6 drive RAID(frame) 10 setup. TheAsus P5Q motherboard supplies 6 SATA ports, plus I added a 2 port ATA card for the (older) boot drive and it's mirror.Andreas, In your description, I did not see how you kept the dk(4) units from attaching to different GPT partitions? Do you have a custom kernelconfiguration that attaches each dkN to a particular GUID, or does yourraidframe configuration somehow do that?
For the low level components, I rely on RAIDframe's auto-configure capability, which finds the components needed to configure each raid set. Intentional (and some real) drive failures and re-ordering, including physically disconnecting one drive in each RAID1 pair and rebooting, always left me with the RAID1 sets correctly auto-configured.
The RAID0 set built on top of the RAID1s cannot be set to auto configure - I am watching the discussion on tech-kern on "RAIDframe nested autoconfiguration". Currently, if the RAID1 sets would be configured in a different order the configuration of the RAID0 set will fail with "Column out of alignment for: /dev/dkXXX". For now, I expect to have to fix that manually, although I have not seen it in over 80 boots. In case it does start happening with annoying frequency, I'll do the configuration of the RAID0 set with a script that identifies the components by their GPT GUID.
--
aew