Okay, I've been working on getting a RAID1 (mirror) setup running on an empeg, and I've got everything running, except I have one major problem that I was hoping to get some feedback on.

It turns out that the RAID arrays themselves don't like being simply turned off. It doesn't affect the data on the drive or anything; it all still runs fine. However, it has to rebuild the parity on the arrays when they're not shut down properly. Unfortuntely, this rebuild takes about 90 minutes to complete on just a pair of 10MB drives. And if a drive dies before the parity is recomputed, it's dead (supposedly -- I haven't actually tested that statement).

I've tried placing a call to the array stop function in the kernel's poweroff procedure, but it doesn't seem to work, despite the fact that it definitely returns (probably because the filesystems are still mounted, but I haven't checked that yet, either).

So, does anyone have any bright ideas?

So, does anyone have any bright ideas?

I don't know if this is even possible under linux's software raid, but it occurs to me that RAID is useless for normal read only operation. It would save drive wear and tear and maybe help this problem if you only mounted the RAID for read/write mounts (or if a drive fails). For read-only situations only mount one half of the mirror bypassing the RAID software entirely and keeping one drive spun down.

-Mike

I don't understand why RAID is useless for read-only operation. If I have a mirror and one half fails, then it starts using the other half. Otherwise, I have to write some logic to figure out which half has failed, which is pretty much like reinventing the wheel. Not to mention how I would go about finding which half failed when intially mounting the root filesystem.

I don't understand why RAID is useless for read-only operation.

Well, maybe useless was too strong a word, but if it was my harddrives I would rather have the second one spun down than getting worn out when there is another perfectly good copy of the data on a live drive. Is there a reason to have the second drive spinning unless the first drive fails (we probably don't care about the minor increase in read performance)? So if the first drive does fail the empeg is going to crash at which point you pull it out of the dash and push it back in and it starts using the second drive instead. Granted it would make the driver more complicated.

Not to mention how I would go about finding which half failed when intially mounting the root filesystem.

Hmmm, so you are planning on doing root on RAID also? I had also assumed you where just going to mirror the music partitions since the other partitions rarely change. Again that would require a reboot if a drive crashed though. I can understand the appeal of having true RAID where the system just keeps running in the event of a drive failure, it just seems like for the empeg an offline mirror might be easier and a better use of resources.

-Mike

Well, I wouldn't use it, but there were one or two people who were interested in mirroring, and I offered to try to get it working because I wanted to work on something.

Unfortunately, I don't believe that Linux RAID can run half a RAID mirror without going through the RAID driver because it stores metadata in the actual partition rather than out of band, like, for example, Solaris RAIDs.

And the reason that the RAID needs to be on all partitions is because if a drive were to fail, then the root partition could go away, which makes the fact that the music is still there irrelevant, as the requester's point was that it would suck to be on a trip and have the empeg go tits up.