I just set up a RAID-5 composed of 3 new Western Digital USB hard drives, each 1TB large. The three drives are connected to the system through a USB hub. The total size of the array is 2TB, formatted as ext3 (largefile4).

I know this USB raid is slow (about 15MB/s slow...), but we ran out of space, SATA ports, and USB ports in the system unit. That's what we get for having to run a server on a budget. We are planning to use the array for semi-long-term storage of large files (DV-AVI files), and decided that sacrificing speed for the redundancy was alright.

Anyway, I moved a large file into the array. Later, while reading the file, I got a read error halfway into the file. I thought it might have been a freak incident as the array was still being resynced when I formatted and copied files in. Also, /proc/mdstat said that all three drives are fine, so I reformatted it again.

I had just started to copy files into the newly formatted array when two of this kind of message pops up on my tty:

I'm starting to suspect the "freak incident" was not a random incident... Does anyone know if anything else could be causing this? (besides hardware failure?)

Auch?


I'm sure you know more about RAID than I do but something tells me that's a bad thing to do...


Certain kernels with >= 4G RAM, too much DMA pressure in other kernels, sata SiliconImage (doesn't apply here but SI is worthless crap and as such deserves mentioning always), bad controller chips (usually cheap no brand encasings)?.. The LKML archives are littered with "ext3_new_block: Allocating block in system zone" cases.

From the little I understand, and from the howtos I've read, the array is supposed to be usable as soon as it is created, and is supposed to still be able to be used in a degraded condition. But I also tried letting it resync first before using it with the same amount of success. (none)

The machine only has 2GB of RAM. The casing came with the drive I bought (WD Elements).

I might have found another lead though.

I was stacking four of the hard drives on top of each other. (Only three of them are in the raid.) After noticing that the lone drive that wasn't in the raid was also showing signs of trouble, I separated them and it made an almost instant recovery. I'll be trying the raid again soon and see if maybe I was just overheating the drives.

Does anyone know if overheating can indeed cause this kind of trouble?

Just an update: I noticed that the data corruption only started *after* the raid was synced. I did an echo check > sync_action and checked mismatch_cnt at regular intervals - the mismatch count kept going up. Seems as though the resync did not work right.

Now I'm testing the drives individually rule out a single hard drive failure.

If anyone has any idea on what I could be doing wrong, please post a reply...