View Full Version : [ubuntu] RAID Design and Kernel

August 17th, 2010, 02:00 AM
Dear Experts:

I have been running ubuntu 10.04 for a while now. The boot drives are software RAID-1 mirrored (mdadm). Interestingly, they nicely mount at bootup and are rather stable AS LONG AS I use the kernel that came with the original ubuntu-server 10.04 release, which is 2.6.32-21. as soon as I try one of the newer kernels pushed via the update, such as the -22, -23, or -24 kernels, the drives are no longer seen. (even the /dev/by-uuid hierarchy disappears, which means that my uuid identified array no longer mounts.)

well, I now have the chance to rebuild my server, because the CPU water-cooler leaked. not fun.

I want to use this opportunity to rebuild my system. I would like some opinions and advice on what a solid configuration is.

I am thinking of building a new RAID array. Safety and simplicity are first and foremost. capacity is unimportant---1GB is enough. so, I am thinking of building a software RAID-0 array with 4 drives. a 1GB drive costs about $50/drive, so we are talking $200. Reading speed is important. Writing speed, oh well, not as important. I presume the read speed of a 4*RAID-1 system will be about 3-4 times that of an individual drive, but the write speed will only be about the same as that of one individual drive.

I presume the recommendation for file system is still ext4. btrfs is close, but no cigar yet. correct?

now, is there a way to get around the weird ubuntu-server mdadm issues?

should I use LVM? again, reading speed and reliability (both hardware crashes and new kernel compatibility) matter. is LVM on its way out now that btrfs is on its way in, so I may as well just wait until btrfs does it?

(is btrfs in all ubuntu server and desktop kernels, so using it would guarantee that any kernel would understand a RAID-1 array controlled by it?)

any other advice?


(It would be nice if the linux RAID-1 were smart enough to do "lazy writes"---as soon as one of the drives has a good copy, the others can be delayed for a few ms . they all should be in themselves coherent, but they don't have to be in exactly the same coherence state. if one drive dies, at most, I should lose a few ms worth of writes. this should improve RAID-1 writing speed.)