I'm experiencing some problems with my raid5 array after upgrading from ubuntu 6.06 to 8.04.

If I transfer large amounts of data to it (for example over FTP or SMB) it stops working properly. Small writes still work, but I suspect they're just being cached. A sync hangs, and can't abort it with ctrl-c. The load creeps up slowly, and usually stops around 30-40. Writes to system disk (80gb sata on a different controller) seem to also be affected.

Most of the time I don't see anything in dmesg, top, ps x or /var/log/messages. I have tried to upgrade kernel to 2.6.24-17-server, as that fixed a similar issue for one at IRC, but still the same.

Basic info:
raid5 array, 3tb, 7 hdd's
Server is 64bit, upgraded from 6.06 to 8.04
XFS filesystem
SATA cards are 2x : Promise Technology, Inc. PDC40718 (SATA 300 TX4) (rev 02)

This worked perfectly in 6.06 server. Upgrade was done via the do-release-upgrade tool.

I have on two occations seen page allocation failures in dmesg following these problems, but I think those are symptoms and not the cause.
[ 1721.538312] swapper: page allocation failure. order:1, mode:0x4020
[ 1721.538320] Pid: 0, comm: swapper Not tainted 2.6.24-17-server #1

As far as I can tell, the problem lies in the kernel, but I'm unsure of where. It can be the XFS filesystem, the raid system, or the basic disk I/O. The kernel is tight lipped as to where it happens.. Anyone got any ideas?