I'm not sure that this is exactly RAID 0+1, but I believe that's what it is called. What I am wanting to do is the following:

sda1 + sdb1 = md0 (RAID 0)
md0 + sdc1 = md1 (RAID 1)

then, I have some space left at sdb2 and sdc2 which I have other non-raid plans for.

Can this be done, or does the software RAID configuration prevent you from choosing a raid array (md0) as one of the disks for another raid array (md1)? If it does prevent you from doing this, is there a workaround?

Any info would be appreciated.

Thanks in advance.

you can get the raid 0 effectively by using lvm, which is going to be a lot more flexible than real raid 0, but essentially you should be able to layer raid. i'd still vote for lvm though.

What advantage would using lvm have?


[edit]After reading up on LVM, I am very impressed with the versatility that LVM provides (at least in theory). However, I am still interested in the performance boost of RAID 0. Can an LVM volume be made to stripe rather than concatenate? or better yet, can I create a an LVM volume out of RAID devices? This possibility is intriguing to me. Thanks for pointing out a technology I wasnt aware of.

LVM uses a "stride" setting to say how to place data. you can certainly turn any valid file system into an lvm partition, so yes you can take two raid-1's and LVM them then into a larger raid device. and then whenever you wanted to, take another pair of drives, raid-1 them, and add them to the logical volume.

also there is absolutely minimal impact of the lvm "layer" as in operation there is simply a translation map held in memory which converts logical space to physical space. some tests, by Novell i think, even showed a fractional performance increase.