Today I found this:

kernel: end_request: I/O error, dev 03:09 (hda), sector 44564575
Nov 11 04:02:23 rev kernel: EXT2-fs error (device ide0(3,9)): ext2_read_inode: unable to read inode block - inode=2774601, block=5570571
kernel: hda: read_intr: status=0x59 { DriveReady SeekComplete DataRequest Error }
kernel: hda: read_intr: error=0x40 { UncorrectableError }, LBAsect=52918438, sector=44564575
kernel: end_request: I/O error, dev 03:09 (hda), sector 44564575
kernel: EXT2-fs error (device ide0(3,9)): ext2_read_inode: unable to read inode block - inode=2774604, block=5570571
kernel: hda: read_intr: status=0x59 { DriveReady SeekComplete DataRequest Error }
kernel: hda: read_intr: error=0x40 { UncorrectableError }, LBAsect=52918438, sector=44564575
kernel: end_request: I/O error, dev 03:09 (hda), sector 44564575
etc.

It's a disk error or a FS error ? UDMA-cable problem? Configuration? Kernel problem? What can I do to fix it ?

Replace HDD?

Any suggestion ?

Seems like a filesystem problem. Of course you've run e2fsck and it didn't correct the problem, or it happened again, then you might have some bad sectors on the disk.

You can run e2fsck -c to put the bad blocks in the bad block inode so that they aren't used.

You might just want to do a man on e2fsck.

To do this remotely, you'll probably want to add the line to your rc.sysinit file, or touch /forcefsck or do a shutdown -F (if the appropriate lines are in your rc.sysinit script), and modify the parameters that fsck runs with in your rc.sysinit.

Of course you might want to schedule some downtime to do this, if you're hosting anyone. Let them know it's going to be down for awhile, depending on the size of the disk.

Have you checked option "Use PCI DMA by default when available" in kernel configuration? Also you could try to use program "hdparm"