We are setting up a SATABeast with 42 1tb disks. With 2 hot spares, this is "40tb" of raw storage. With a couple 20-disk RAID6 sets, it's 18,002,478,366,720 bytes = 16,766gibytes. Problem: RHEL/CentOS 5.1 & 5.2's ext3 filesystem supports slightly less than 8tibytes. RHEL 5.3 includes ext4dev, meaning Red Hat still considers ext4 unstable. CentOS is working on 5.3, but doesn't seem close to release.

The SATABeast is very heavy (151 lbs with dual controllers and 42 drives), so naturally you put it at the bottom of the rack. But there are a couple problems:

  1. All the health & status indicators are on the bottom of the bezel, so you cannot see them without a mirror on the floor.
  2. The drives aren't hot swap. It's even worse than the X4500, with its dire warnings about leaving the cover off for over 60sec to replace a drive -- to get a drive out of the SATABeast, we have to unmount and shut it down, remove two screws and the front bezel, and then remove the failed drive -- they provide a special tool to lift the drive out!
  3. The SATABeast has a serial console (in our dual-controller unit, it's the lower serial port on Controller 0 -- the upper serial port on Controller 1 is inactive the upper controller is #0, and apparently controller #0's GE port #0 is the only one with the administrative web server the administrative GUI is available on GE port #0, but not port #1, of each controller, with preassigned static IPs (10.11.12.13 & 10.11.12.14) -- GE ports #1 are only for iSCSI, which we don't use; both serial ports are usable). Since we installed it last week, the serial port console has failed 3 times, and I have had to reboot to get it back. Yes, I upgraded the firmware.
  4. The firmware is not available for download -- you have to ask support to email it.
  5. Nexsan provides a Mac GUI tool for managing SATABeasts, but it only works via autodiscovery -- it cannot see our SATABeast (in another subnet), and there's no way to specify the IP. Lame.

On the other hand, the GUI is generally well designed (although a bit overly complicated -- there are 11 left-hand menu items, and buttons within the main UI tend to jump into the wrong top-level item, which makes operations such as changing LUN mappings harder than they should be). Also, it's pathetic that Nexsan charges $250 for an SSL certificate, and doesn't let you BYOC.

But it looks like we don't really need RHEL 5.3's ext4 support anyway:

NOTE: Although very large fileystems are on ext4's feature list, current e2fsprogs currently still limits the filesystem size to 2^32 blocks (16T for a 4k block filesystem). Filesystems larger than 16T is one of the very next high-priority features to complete for ext4.