You'll be able to import this VirtualBox image directly into a running
instance of the GUI, assuming you're using vbox version 3 or above.

There are two versions of the kernel to choose from, the default being
the latest 2.6.30.5.  Optionally, there's a stock 2.6.18 provided with
CentOS 5.3.  Userland is a very recent update of the stock CentOS 5.3.

When the vm is online and available, you'll be able to log into the
console with the following credentials:

  username: root
  password: password

Within the virtual machine, you'll observe that /dev/sda is the XFS
drive, mounted at /data and formatted with 'mkfs.xfs -s size=4k /dev/sda.'

To begin the test, use your host operating system (or any spare machine
that has NFS client utilities) to begin a continuous write of data, as such:

$ sudo mkdir /mnt/test
$ sudo mount <IP of virtual machine>:/data /mnt/test
$ while true
    do
    dd if=/dev/zero of=/mnt/test/foo.bin bs=1M count=512
done

Within the virtual machine, uncomment root's cronjob to reset the machine
every 5 minutes:

# crontab -e
*/5 * * * * /root/xfs/die.sh

Wait until the xfs partition can't be mounted on boot to see problem.  We
typically see the error after merely several reoots, but it's been known
to take several hours.   Have a beer!  Note that the cronjob will not reboot
the virtual machine if the xfs partition isn't mounted.  So if the VM
isn't rebooting every 5 minutes, that is another indication that the
failure has occured.

When you detect a failure, log into the VM and observe that /dev/sda
isn't mounted and xfs repair fails unless you add the -L parameter:

# xfs_repair /dev/sda

Remember to comment out that crontab entry so you don't re-kill yourself.