PCI SATA / RAID Issues

[[R4A]Bigman]

My housemate and I are currently in the process of rebuilding our fileserver with some lurvly new 500GB SATA disks. First things first, we used Solaris (for the first time ever) and constructed the thing, and after a lot of faffing around got it working. Using ZFS and RAID-Z we had the disks working. However, on doing some copying to and fro, ZFS reported I/O errors - indicating that the data/parity was inconsistent and couldn't be rebuilt from checksum - basically a big bad sign.

Putting this off to the fact we didn't really know what we were doing in Solaris, we got ubuntu-server running and used software RAID-5 (mdadm). However, every time we boot and start taxing the hard drives, like rebuilding the array for example, we end up with kernel reports that drives are failing to respond or slow to respond, soft retries and hard retries etc. The errors come from all the disks and all the controller cards we try in any configuration.

Hardware:
AMD XP1700+
384MB PC2100 RAM
Abit NF7-S Mobo
4x 500GB WD AAKS disks
Silicon Image SI3112 (yellow card)
Silicon Image SI3112 (black card)
Silicon Image SI3114
Silicon Image SI3112 (on board)

We've tried every combination of the cards in or out of the machine, with the onboard enabled and disabled and we always get some nasty errors. The one which appears to fail silently is the SI3114, where we don't get any errors thrown back at the OS, but instead we end up with data being corrupted (different MD5sums being reported once copied).

I've really grown a dislike for Silicon Image based cards over the years, but I'm not going to be too quick to jump to the conclusion that they are the source of all our problems. If anyone can suggest anything at all which might help, that would be absolutely excellent.

All comments, suggestions or whatever are all very welcome and appreciated

[arthurleung]

I think Silicon Image's driver is quite crap on anything other than Windows, otherwise I find them quite decent for basic disk connectivity. (However I wouldn't trust their software RAID stack)

Have you consider the possibility that your NF2 chipset is the culprit?
Try updating the BIOS.

I've had severe data corruption on my NF4 chipset when I load the PCI bus too much, until I updated the BIOS.

[[R4A]Bigman]

We had to flash the Silicon Image BIOSes to the latest ones to get them to work at all under Solaris - I believe they just appear to be an IDE device to the OS now, so driver issues shouldn't be an issue with them.

There is the possibility that it is my NF2 chipset, so I will get on flashing that to the latest version asap.

We stress tested the 4 disks through a NF4 onboard SATA, and we didn't have any data corruption then, suggesting the disks themselves are fine. The next step is running them together through the various SI cards on this NF4 system to see if we can get the data to corrupt that way (thus determining whether its the card or the system to blame).

Thanks for the suggestion

[[R4A]Bigman]

Well, either we got lucky with something (doubtful) or flashing the Mobo BIOS actually did work.

The 4-port card appears to be working perfectly now under both ubuntu-server and solaris. Before we trust it with our data we are going to do some more vigorous testing (such as copying a whole load of data to it, leaving it for a while and checking MD5 sums).

Thanks very much for the helps