OK, I haven't found this issue mentioned in this fourm, but I could have missed it, if I did just link it related and pls don't flame.

System info:
gentoo-2.6.11 kernel no patches. (Si support compiled in, raid support compiled in)
1.4 GHz athlon on a K266 MB with 512MB ram

Background:
A while ago I decided to try software raid to but together a bunch of 250 GB harddrives. My goal was to put just over 1TB under my desk. I wanted to string together 6 250 GB Western Digital SATA drives in raid 5 for a non wallet destroying price(software raid). I did some research to make sure the drives were not blacklisted. Then I bought two 4 port SATA controllers.
http://www.syba.com/product/43/02/05/index.html (SATA card)
http://www.wdc.com/en/products/Products.asp?DriveID=59 (Drives)

I believe the controllers are the SI 3114 chipset.

How the problem Exhibits itself:
When the array gets busy drives get kicked out of the raid array.
Access to the kicked out drive(s) results in a timeout for what ever utility I use to get stats on it. (mdadm, fdisk, mount, anything that accesses the device)
Anything that uses the drive is extremely slow or does not work.
-fdisk read the partition information incorrectly and cannot modify anything
-raid becomes extremely slow
-If I try to wipe the kicked out disk nothing happens when I do a mkfs
After time other I/O systems start to fail.
Upon reboot the raid array does not unmount and I get an error code from the drive(s) that got kicked out. (I don't have the code in front of me but I will update later)

I found a similar issue with an earlier model here:
http://forums.gentoo.org/viewtopic-t...ata+4port.html

Why I think this is a driver problem:
I yank all the drives from the raid array and made them stand alone vanilla sata drives. I error checked the drives and found no problems.

Other observations:
I have a total of 8 of these drives, the bigger I make the array the faster disks get kicked out.
There is no bias to any single drive getting kicked out.
If the array is broken and rebuilt I have no problem until the array gets busy again.
If the array is > 6 disks a dsk get kicked out about 15 minutes after the array is created.

My WAG:
Something is causeing an extreme smount of PCI traffic. By extreme I mean waaaaay more than it should. This causes some type of internal timeout. The drive gets marked bad by the md driver(rightfully so since it is non-reponsive) and kicked out. The kernel can't recover and the badness spreads to other I/O systems.

I'll link if I find a solution, if you know of a solution pls pls pls link me to it.

-Darkseer

This is related I think:
http://www.thisishull.net/archive/in...p/t-21928.html

Sorry, I do not have any suggests that will fix your problems. Since you are in the beginning of setting up the file server, have you looked at Highpoint RocketRAID 1540 or Highpoint RocketRAID 1640. I have a Highpoint controller (HPT370) in two systems and they have not give me any problems on very long (+4 hour) video recordings. If I had to setup a file server that reaches up to a terabyte or more, I would use Highpoint controllers when price and reliability is an issue. The kernel does have support up to HPT374 which Highpoint RocketRAID 1640 and Highpoint RocketRAID 1540 uses.

RAID 5 is not the best and cheap to use for personal file servers. The best and the cheapest is to setup multiple RAID 1 (mirroring) using two hard drives for each array and setup it up in an LVM or EVMS partition. This way you can increase your storage when you need to. You will still get reduntancy. The processor will work less when using RAID 1 and LVM or EVMS instead of RAID 5. Software RAID 5 is a resource hog.

The practical implementation of raid is all still a learning experience to me. I'll look into the raid controllers you recomended. I have issues with volume management though, EVMS in particular. When one of my disks had a fake failure it crashed. Then during reboot it crashed again halting the boot process. I needed to go single user to take it out of the boot sequence. While I am attracted to the flexability it offers, I think there are still some issues that need to be ironed out. Thanks for your recomendation. I'm going to troll the kernel mailing lists to see if they've mentioned anything about this.

After looking through other mailing groups I've found that the Silicon image drivers are still sorta beta. I switched to two promise tx4 controllers and am testing them now. So I guess you get what you pay for. Promise controllers were $52 each and the Si controllers were $20 each.

-Darkseer

Promise controllers will also have similar problems. I have read that Promise controllers have problems with DMA. Some people spend the money for Windows 2003 Server because Silicon Image and Promise controllers have problems in Linux. Hopefully, I do not have to say "I told you so".